snapshot of https://github.com/ggml-org/whisper.cpp.git @ da54572229bcf64ba367d96c7ef15770376c4280 (2026-09-15)

2026-09-15 17:23:35 +05:00
commit 948ef47106
1955 changed files with 719115 additions and 0 deletions
--- a/tests/.gitignore
+++ b/tests/.gitignore
@@ -0,0 +1,3 @@
+*.wav
+*.ogg
+*.wav.txt
--- a/tests/CMakeLists.txt
+++ b/tests/CMakeLists.txt
@@ -0,0 +1,196 @@
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+
+if (EMSCRIPTEN)
+    #
+    # test-whisper-js
+
+    set(TEST_TARGET test-whisper-js)
+
+    add_test(NAME ${TEST_TARGET}
+        COMMAND node test-whisper.js --experimental-wasm-threads
+        WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
+        )
+
+    return()
+endif()
+
+set(TEST_TARGET test-whisper-cli-tiny)
+add_test(NAME ${TEST_TARGET}
+    COMMAND $<TARGET_FILE:whisper-cli>
+    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-tiny.bin -l fr
+    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
+set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "tiny;gh")
+
+set(TEST_TARGET test-whisper-cli-tiny.en)
+add_test(NAME ${TEST_TARGET}
+    COMMAND $<TARGET_FILE:whisper-cli>
+    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-tiny.en.bin
+    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
+set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "tiny;en")
+
+set(TEST_TARGET test-whisper-cli-base)
+add_test(NAME ${TEST_TARGET}
+    COMMAND $<TARGET_FILE:whisper-cli>
+    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-base.bin -l fr
+    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
+set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "base")
+
+set(TEST_TARGET test-whisper-cli-base.en)
+add_test(NAME ${TEST_TARGET}
+    COMMAND $<TARGET_FILE:whisper-cli>
+    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-base.en.bin
+    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
+set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "base;en")
+
+set(TEST_TARGET test-whisper-cli-small)
+add_test(NAME ${TEST_TARGET}
+    COMMAND $<TARGET_FILE:whisper-cli>
+    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-small.bin -l fr
+    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
+set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "small")
+
+set(TEST_TARGET test-whisper-cli-small.en)
+add_test(NAME ${TEST_TARGET}
+    COMMAND $<TARGET_FILE:whisper-cli>
+    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-small.en.bin
+    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
+set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "small;en")
+
+set(TEST_TARGET test-whisper-cli-medium)
+add_test(NAME ${TEST_TARGET}
+    COMMAND $<TARGET_FILE:whisper-cli>
+    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-medium.bin -l fr
+    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
+set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "medium")
+
+set(TEST_TARGET test-whisper-cli-medium.en)
+add_test(NAME ${TEST_TARGET}
+    COMMAND $<TARGET_FILE:whisper-cli>
+    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-medium.en.bin
+    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
+set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "medium;en")
+
+set(TEST_TARGET test-whisper-cli-large)
+add_test(NAME ${TEST_TARGET}
+    COMMAND $<TARGET_FILE:whisper-cli>
+    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-large.bin
+    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
+set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "large")
+
+if (WHISPER_COMMON_FFMPEG)
+    set(TEST_TARGET test-whisper-cli-tiny-mp3)
+    # Check with reviewers: any way to check the output transcription via ctest (diff, ...)?
+    add_test(NAME ${TEST_TARGET}
+      COMMAND $<TARGET_FILE:whisper-cli>
+      -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-tiny.en.bin
+      -f ${PROJECT_SOURCE_DIR}/samples/jfk.mp3)
+    set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "tiny;mp3")
+endif()
+
+# UTF-8 helper unit test
+set(UTF8_TEST test-common-utf8)
+add_executable(${UTF8_TEST} ${UTF8_TEST}.cpp)
+target_include_directories(${UTF8_TEST} PRIVATE ../examples)
+target_link_libraries(${UTF8_TEST} PRIVATE common)
+add_test(NAME ${UTF8_TEST} COMMAND ${UTF8_TEST})
+set_tests_properties(${UTF8_TEST} PROPERTIES LABELS "unit")
+
+set(BUFFER_LOADER_TEST test-whisper-buffer-loader)
+add_executable(${BUFFER_LOADER_TEST} ${BUFFER_LOADER_TEST}.cpp)
+target_include_directories(${BUFFER_LOADER_TEST} PRIVATE ../include ../ggml/include ../examples)
+target_link_libraries(${BUFFER_LOADER_TEST} PRIVATE common)
+add_test(NAME ${BUFFER_LOADER_TEST} COMMAND ${BUFFER_LOADER_TEST})
+set_tests_properties(${BUFFER_LOADER_TEST} PROPERTIES LABELS "unit;gh")
+
+# whisper_full() with n_samples == 0 must not read an uninitialized mel (#3978)
+set(ZERO_SAMPLES_TEST test-whisper-zero-samples)
+add_executable(${ZERO_SAMPLES_TEST} ${ZERO_SAMPLES_TEST}.cpp)
+target_include_directories(${ZERO_SAMPLES_TEST} PRIVATE ../include ../ggml/include ../examples)
+target_link_libraries(${ZERO_SAMPLES_TEST} PRIVATE common)
+target_compile_definitions(${ZERO_SAMPLES_TEST} PRIVATE
+    WHISPER_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-ggml-tiny.bin")
+add_test(NAME ${ZERO_SAMPLES_TEST} COMMAND ${ZERO_SAMPLES_TEST})
+set_tests_properties(${ZERO_SAMPLES_TEST} PROPERTIES LABELS "tiny;gh")
+
+# VAD test tests VAD in isolation
+set(VAD_TEST test-vad)
+add_executable(${VAD_TEST} ${VAD_TEST}.cpp)
+target_include_directories(${VAD_TEST} PRIVATE ../include ../ggml/include ../examples)
+target_link_libraries(${VAD_TEST} PRIVATE common)
+target_compile_definitions(${VAD_TEST} PRIVATE
+    VAD_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-silero-v6.2.0-ggml.bin"
+    SAMPLE_PATH="${PROJECT_SOURCE_DIR}/samples/jfk.wav")
+add_test(NAME ${VAD_TEST} COMMAND ${VAD_TEST})
+set_tests_properties(${VAD_TEST} PROPERTIES LABELS "unit")
+
+# VAD test full uses whisper_full with VAD enabled
+set(VAD_TEST test-vad-full)
+add_executable(${VAD_TEST} ${VAD_TEST}.cpp)
+target_include_directories(${VAD_TEST} PRIVATE ../include ../ggml/include ../examples)
+target_link_libraries(${VAD_TEST} PRIVATE common)
+target_compile_definitions(${VAD_TEST} PRIVATE
+    WHISPER_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/ggml-base.en.bin"
+    VAD_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-silero-v6.2.0-ggml.bin"
+    SAMPLE_PATH="${PROJECT_SOURCE_DIR}/samples/jfk.wav")
+add_test(NAME ${VAD_TEST} COMMAND ${VAD_TEST})
+set_tests_properties(${VAD_TEST} PROPERTIES LABELS "base;en")
+
+# Parakeet model loading test
+set(PARAKEET_TEST test-parakeet)
+add_executable(${PARAKEET_TEST} ${PARAKEET_TEST}.cpp)
+target_include_directories(${PARAKEET_TEST} PRIVATE ../include ../ggml/include ../examples)
+target_link_libraries(${PARAKEET_TEST} PRIVATE parakeet common)
+target_compile_definitions(${PARAKEET_TEST} PRIVATE
+    PARAKEET_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-ggml-parakeet-tdt.bin"
+    PARAKEET_BAD_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-ggml-parakeet-tdt-bad-nfft0.bin"
+    SAMPLE_PATH="${PROJECT_SOURCE_DIR}/samples/jfk.wav")
+add_test(NAME ${PARAKEET_TEST} COMMAND ${PARAKEET_TEST})
+set_tests_properties(${PARAKEET_TEST} PROPERTIES LABELS "parakeet;gh")
+
+# The following parakeet test require a real ggml-parakeet-tdt model to have
+# been converted or downloaded:
+# $ hf download danbev/parakeet parakeet-tdt-0.6b-v3-f32.bin --local-dir models
+#
+# And also required more audio samples that are shipped by default. These can
+# downloaded by running:
+# $ make samples
+function(add_parakeet_transcription_test TEST_TARGET TEST_SOURCE SAMPLE_PATH EXPECTED_TRANSCRIPTION_PATH)
+    set(TRANSCRIPTION_SIMILARITY_THRESHOLD "1.0")
+    if (ARGC GREATER 4)
+        set(TRANSCRIPTION_SIMILARITY_THRESHOLD "${ARGV4}")
+    endif()
+
+    add_executable(${TEST_TARGET} ${TEST_SOURCE})
+    target_include_directories(${TEST_TARGET} PRIVATE ../include ../ggml/include ../examples)
+    target_link_libraries(${TEST_TARGET} PRIVATE parakeet common)
+    target_compile_definitions(${TEST_TARGET} PRIVATE
+        PARAKEET_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/ggml-parakeet-tdt-0.6b-v3-f32.bin"
+        SAMPLE_PATH="${PROJECT_SOURCE_DIR}/${SAMPLE_PATH}"
+        EXPECTED_TRANSCRIPTION_PATH="${PROJECT_SOURCE_DIR}/${EXPECTED_TRANSCRIPTION_PATH}"
+        TRANSCRIPTION_SIMILARITY_THRESHOLD=${TRANSCRIPTION_SIMILARITY_THRESHOLD})
+
+    add_custom_target(run-${TEST_TARGET}
+        COMMAND $<TARGET_FILE:${TEST_TARGET}>
+        DEPENDS ${TEST_TARGET}
+        WORKING_DIRECTORY ${PROJECT_BINARY_DIR})
+endfunction()
+
+add_parakeet_transcription_test(
+    test-parakeet-full-jfk
+    test-parakeet-full.cpp
+    samples/jfk.wav
+    tests/parakeet-expected-jfk-output.txt)
+
+add_parakeet_transcription_test(
+    test-parakeet-full-gb1
+    test-parakeet-full.cpp
+    samples/gb1.wav
+    tests/parakeet-expected-gb1-output.txt)
+
+add_parakeet_transcription_test(
+    test-parakeet-full-diffusion
+    test-parakeet-full.cpp
+    samples/diffusion2023-07-03.flac
+    tests/parakeet-expected-diffusion-output.txt
+    0.95)
--- a/tests/earnings21/.gitignore
+++ b/tests/earnings21/.gitignore
@@ -0,0 +1,6 @@
+__pycache__
+*.tar.gz
+*.txt
+eval.conf
+venv
+speech-datasets
--- a/tests/earnings21/Makefile
+++ b/tests/earnings21/Makefile
@@ -0,0 +1,16 @@
+GIT_URL = https://github.com/revdotcom/speech-datasets
+
+all: eval
+
+eval:
+	$(MAKE) -f eval.mk
+
+clean:
+	$(MAKE) -f eval.mk clean
+
+get-audio:
+	git clone --depth 1  --filter=blob:none  --sparse $(GIT_URL)
+	git -C speech-datasets sparse-checkout init --cone
+	git -C speech-datasets sparse-checkout set earnings21
+
+.PHONY: all eval clean get-audio
--- a/tests/earnings21/README.md
+++ b/tests/earnings21/README.md
@@ -0,0 +1,87 @@
+# whisper.cpp/tests/earnings21
+
+[Earnings-21](https://arxiv.org/abs/2104.11348) is a real-world benchmark
+dataset that contains 39-hours of long-form English speech, sourced from
+public earning calls.
+
+This directory contains a set of scripts to evaluate the performance of
+whisper.cpp on Earnings-21 corpus.
+
+## Quick Start
+
+1. (Pre-requirement) Compile `whisper-cli` and prepare the Whisper
+   model in `ggml` format.
+
+   ```
+   $ # Execute the commands below in the project root dir.
+   $ cmake -B build
+   $ cmake --build build --config Release
+   $ ./models/download-ggml-model.sh tiny
+   ```
+
+   Consult [whisper.cpp/README.md](../../README.md) for more details.
+
+2. Download the audio files.
+
+   ```
+   $ make get-audio
+   ```
+
+3. Set up the environment to compute WER score.
+
+   ```
+   $ pip install -r requirements.txt
+   ```
+
+   For example, if you use `virtualenv`, you can set up it as follows:
+
+   ```
+   $ python3 -m venv venv
+   $ . venv/bin/activate
+   $ pip install -r requirements.txt
+   ```
+
+4. Run the benchmark test.
+
+   ```
+   $ make
+   ```
+
+## How-to guides
+
+### How to change the inference parameters
+
+Create `eval.conf` and override variables.
+
+```
+WHISPER_MODEL = large-v3-turbo
+WHISPER_FLAGS = --no-prints --threads 8 --language en --output-txt
+```
+
+Check out `eval.mk` for more details.
+
+### How to perform the benchmark test on a 10-hour subset
+
+Earnings-21 provides a small but representative subset (approximately
+10-hour audio data) to evaluate ASR systems quickly.
+
+To switch to the subset, create `eval.conf` and add the following line:
+
+```
+EARNINGS21_EVAL10 = yes
+```
+
+### How to run the benchmark test using VAD
+
+First, you need to download a VAD model:
+
+```
+$ # Execute the commands below in the project root dir.
+$ ./models/download-vad-model.sh silero-v6.2.0
+```
+
+Create `eval.conf` with the following content:
+
+```
+WHISPER_FLAGS = --no-prints --language en --output-txt --vad --vad-model ../../models/ggml-silero-v6.2.0.bin
+```
--- a/tests/earnings21/eval.mk
+++ b/tests/earnings21/eval.mk
@@ -0,0 +1,58 @@
+PYTHON = python
+
+WHISPER_PREFIX = ../../
+WHISPER_MODEL = tiny
+
+WHISPER_CLI = $(WHISPER_PREFIX)build/bin/whisper-cli
+WHISPER_FLAGS = --no-prints --language en --output-txt
+
+# You can create eval.conf to override the WHISPER_* variables
+# defined above.
+-include eval.conf
+
+# Add  `EARNINGS21_EVAL10 = yes` to eval.conf to switch to a
+# 10-hour subset. See "speech-datasets/earnings21/README.md" for
+# more details about this subset.
+ifdef EARNINGS21_EVAL10
+METADATA_CSV = speech-datasets/earnings21/eval10-file-metadata.csv
+AUDIO_SRCS = speech-datasets/earnings21/media/4320211.mp3 \
+             speech-datasets/earnings21/media/4341191.mp3 \
+             speech-datasets/earnings21/media/4346818.mp3 \
+             speech-datasets/earnings21/media/4359971.mp3 \
+             speech-datasets/earnings21/media/4365024.mp3 \
+             speech-datasets/earnings21/media/4366522.mp3 \
+             speech-datasets/earnings21/media/4366893.mp3 \
+             speech-datasets/earnings21/media/4367535.mp3 \
+             speech-datasets/earnings21/media/4383161.mp3 \
+             speech-datasets/earnings21/media/4384964.mp3 \
+             speech-datasets/earnings21/media/4387332.mp3
+else
+METADATA_CSV = speech-datasets/earnings21/earnings21-file-metadata.csv
+AUDIO_SRCS = $(sort $(wildcard speech-datasets/earnings21/media/*.mp3))
+endif
+
+TRANS_TXTS = $(addsuffix .txt, $(AUDIO_SRCS))
+
+# We output the evaluation result to this file.
+DONE = $(WHISPER_MODEL).txt
+
+all: $(DONE)
+
+$(DONE): $(TRANS_TXTS)
+	$(PYTHON) eval.py $(METADATA_CSV) > $@.tmp
+	mv $@.tmp $@
+
+# Note: This task writes to a temporary file first to
+# create the target file atomically.
+%.mp3.txt: %.mp3
+	$(WHISPER_CLI) $(WHISPER_FLAGS) --model $(WHISPER_PREFIX)models/ggml-$(WHISPER_MODEL).bin --file $^ --output-file $^.tmp
+	mv $^.tmp.txt $^.txt
+
+archive:
+	tar -czf $(WHISPER_MODEL).tar.gz --exclude="*.mp3" speech-datasets/earnings21/media $(DONE)
+
+clean:
+	@rm -f $(TRANS_TXTS)
+	@rm -f $(DONE)
+
+.PHONY: all archive clean
--- a/tests/earnings21/eval.py
+++ b/tests/earnings21/eval.py
@@ -0,0 +1,68 @@
+import os
+import sys
+import glob
+import jiwer
+from normalizers import EnglishTextNormalizer
+
+def decode_hypothesis(b):
+    try:
+        # Depending on platforms, Whisper can emit a left double quotation
+        # mark (0x93), which is Microsoft's extension to ASCII. See #3185
+        # for the background.
+        return b.decode('windows-1252')
+    except UnicodeDecodeError:
+        return b.decode('utf-8', errors='ignore')
+
+def get_reference():
+    ref = {}
+    for path in glob.glob("speech-datasets/earnings21/transcripts/nlp_references/*.nlp"):
+        code = os.path.basename(path).replace(".nlp", "")
+        buf = []
+        with open(path) as fp:
+            fp.readline()
+            for line in fp:
+                token = line.split("|", maxsplit=1)[0]
+                buf.append(token)
+            ref[code] = " ".join(buf)
+    return ref
+
+def get_hypothesis():
+    hyp = {}
+    for path in glob.glob("speech-datasets/earnings21/media/*.mp3.txt"):
+        with open(path, 'rb') as fp:
+            text = decode_hypothesis(fp.read()).strip()
+        code = os.path.basename(path).replace(".mp3.txt", "")
+        hyp[code] = text
+    return hyp
+
+def get_codes(metadata_csv):
+    codes = []
+    with open(metadata_csv) as fp:
+        fp.readline()
+        for line in fp:
+            codes.append(line.split(",")[0])
+    return sorted(codes)
+
+def main():
+    if len(sys.argv) < 2:
+        print("Usage: %s METADATA_CSV" % sys.argv[0], file=sys.stderr)
+        return 1
+
+    metadata_csv = sys.argv[1]
+    normalizer = EnglishTextNormalizer()
+
+    ref_orig = get_reference()
+    hyp_orig = get_hypothesis()
+
+    ref_clean = []
+    hyp_clean = []
+
+    for code in get_codes(metadata_csv):
+        ref_clean.append(normalizer(ref_orig[code]))
+        hyp_clean.append(normalizer(hyp_orig[code]))
+
+    wer = jiwer.wer(ref_clean, hyp_clean)
+    print(f"WER: {wer * 100:.2f}%")
+
+if __name__ == "__main__":
+    main()
--- a/tests/earnings21/normalizers/LICENSE
+++ b/tests/earnings21/normalizers/LICENSE
@@ -0,0 +1,25 @@
+Code in this directory is adapted from OpenAI Whisper project
+(https://github.com/openai/whisper) and carries the following
+copyright and license.
+
+    MIT License
+
+    Copyright (c) 2022 OpenAI
+
+    Permission is hereby granted, free of charge, to any person obtaining a copy
+    of this software and associated documentation files (the "Software"), to deal
+    in the Software without restriction, including without limitation the rights
+    to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+    copies of the Software, and to permit persons to whom the Software is
+    furnished to do so, subject to the following conditions:
+
+    The above copyright notice and this permission notice shall be included in all
+    copies or substantial portions of the Software.
+
+    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+    OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+    SOFTWARE.
--- a/tests/earnings21/normalizers/init.py
+++ b/tests/earnings21/normalizers/init.py
@@ -0,0 +1,2 @@
+from .basic import BasicTextNormalizer as BasicTextNormalizer
+from .english import EnglishTextNormalizer as EnglishTextNormalizer
--- a/tests/earnings21/normalizers/basic.py
+++ b/tests/earnings21/normalizers/basic.py
@@ -0,0 +1,80 @@
+import re
+import unicodedata
+
+import regex
+
+# non-ASCII letters that are not separated by "NFKD" normalization
+ADDITIONAL_DIACRITICS = {
+    "œ": "oe",
+    "Œ": "OE",
+    "ø": "o",
+    "Ø": "O",
+    "æ": "ae",
+    "Æ": "AE",
+    "ß": "ss",
+    "ẞ": "SS",
+    "đ": "d",
+    "Đ": "D",
+    "ð": "d",
+    "Ð": "D",
+    "þ": "th",
+    "Þ": "th",
+    "ł": "l",
+    "Ł": "L",
+}
+
+
+def remove_symbols_and_diacritics(s: str, keep=""):
+    """
+    Replace any other markers, symbols, and punctuations with a space,
+    and drop any diacritics (category 'Mn' and some manual mappings)
+    """
+    return "".join(
+        (
+            c
+            if c in keep
+            else (
+                ADDITIONAL_DIACRITICS[c]
+                if c in ADDITIONAL_DIACRITICS
+                else (
+                    ""
+                    if unicodedata.category(c) == "Mn"
+                    else " " if unicodedata.category(c)[0] in "MSP" else c
+                )
+            )
+        )
+        for c in unicodedata.normalize("NFKD", s)
+    )
+
+
+def remove_symbols(s: str):
+    """
+    Replace any other markers, symbols, punctuations with a space, keeping diacritics
+    """
+    return "".join(
+        " " if unicodedata.category(c)[0] in "MSP" else c
+        for c in unicodedata.normalize("NFKC", s)
+    )
+
+
+class BasicTextNormalizer:
+    def __init__(self, remove_diacritics: bool = False, split_letters: bool = False):
+        self.clean = (
+            remove_symbols_and_diacritics if remove_diacritics else remove_symbols
+        )
+        self.split_letters = split_letters
+
+    def __call__(self, s: str):
+        s = s.lower()
+        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
+        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
+        s = self.clean(s).lower()
+
+        if self.split_letters:
+            s = " ".join(regex.findall(r"\X", s, regex.U))
+
+        s = re.sub(
+            r"\s+", " ", s
+        )  # replace any successive whitespace characters with a space
+
+        return s
--- a/tests/earnings21/normalizers/english.json
+++ b/tests/earnings21/normalizers/english.json
--- a/tests/earnings21/normalizers/english.py
+++ b/tests/earnings21/normalizers/english.py
@@ -0,0 +1,550 @@
+import json
+import os
+import re
+from fractions import Fraction
+from typing import Iterator, List, Match, Optional, Union
+
+from more_itertools import windowed
+
+from .basic import remove_symbols_and_diacritics
+
+
+class EnglishNumberNormalizer:
+    """
+    Convert any spelled-out numbers into arabic numbers, while handling:
+
+    - remove any commas
+    - keep the suffixes such as: `1960s`, `274th`, `32nd`, etc.
+    - spell out currency symbols after the number. e.g. `$20 million` -> `20000000 dollars`
+    - spell out `one` and `ones`
+    - interpret successive single-digit numbers as nominal: `one oh one` -> `101`
+    """
+
+    def __init__(self):
+        super().__init__()
+
+        self.zeros = {"o", "oh", "zero"}
+        self.ones = {
+            name: i
+            for i, name in enumerate(
+                [
+                    "one",
+                    "two",
+                    "three",
+                    "four",
+                    "five",
+                    "six",
+                    "seven",
+                    "eight",
+                    "nine",
+                    "ten",
+                    "eleven",
+                    "twelve",
+                    "thirteen",
+                    "fourteen",
+                    "fifteen",
+                    "sixteen",
+                    "seventeen",
+                    "eighteen",
+                    "nineteen",
+                ],
+                start=1,
+            )
+        }
+        self.ones_plural = {
+            "sixes" if name == "six" else name + "s": (value, "s")
+            for name, value in self.ones.items()
+        }
+        self.ones_ordinal = {
+            "zeroth": (0, "th"),
+            "first": (1, "st"),
+            "second": (2, "nd"),
+            "third": (3, "rd"),
+            "fifth": (5, "th"),
+            "twelfth": (12, "th"),
+            **{
+                name + ("h" if name.endswith("t") else "th"): (value, "th")
+                for name, value in self.ones.items()
+                if value > 3 and value != 5 and value != 12
+            },
+        }
+        self.ones_suffixed = {**self.ones_plural, **self.ones_ordinal}
+
+        self.tens = {
+            "twenty": 20,
+            "thirty": 30,
+            "forty": 40,
+            "fifty": 50,
+            "sixty": 60,
+            "seventy": 70,
+            "eighty": 80,
+            "ninety": 90,
+        }
+        self.tens_plural = {
+            name.replace("y", "ies"): (value, "s") for name, value in self.tens.items()
+        }
+        self.tens_ordinal = {
+            name.replace("y", "ieth"): (value, "th")
+            for name, value in self.tens.items()
+        }
+        self.tens_suffixed = {**self.tens_plural, **self.tens_ordinal}
+
+        self.multipliers = {
+            "hundred": 100,
+            "thousand": 1_000,
+            "million": 1_000_000,
+            "billion": 1_000_000_000,
+            "trillion": 1_000_000_000_000,
+            "quadrillion": 1_000_000_000_000_000,
+            "quintillion": 1_000_000_000_000_000_000,
+            "sextillion": 1_000_000_000_000_000_000_000,
+            "septillion": 1_000_000_000_000_000_000_000_000,
+            "octillion": 1_000_000_000_000_000_000_000_000_000,
+            "nonillion": 1_000_000_000_000_000_000_000_000_000_000,
+            "decillion": 1_000_000_000_000_000_000_000_000_000_000_000,
+        }
+        self.multipliers_plural = {
+            name + "s": (value, "s") for name, value in self.multipliers.items()
+        }
+        self.multipliers_ordinal = {
+            name + "th": (value, "th") for name, value in self.multipliers.items()
+        }
+        self.multipliers_suffixed = {
+            **self.multipliers_plural,
+            **self.multipliers_ordinal,
+        }
+        self.decimals = {*self.ones, *self.tens, *self.zeros}
+
+        self.preceding_prefixers = {
+            "minus": "-",
+            "negative": "-",
+            "plus": "+",
+            "positive": "+",
+        }
+        self.following_prefixers = {
+            "pound": "£",
+            "pounds": "£",
+            "euro": "€",
+            "euros": "€",
+            "dollar": "$",
+            "dollars": "$",
+            "cent": "¢",
+            "cents": "¢",
+        }
+        self.prefixes = set(
+            list(self.preceding_prefixers.values())
+            + list(self.following_prefixers.values())
+        )
+        self.suffixers = {
+            "per": {"cent": "%"},
+            "percent": "%",
+        }
+        self.specials = {"and", "double", "triple", "point"}
+
+        self.words = set(
+            [
+                key
+                for mapping in [
+                    self.zeros,
+                    self.ones,
+                    self.ones_suffixed,
+                    self.tens,
+                    self.tens_suffixed,
+                    self.multipliers,
+                    self.multipliers_suffixed,
+                    self.preceding_prefixers,
+                    self.following_prefixers,
+                    self.suffixers,
+                    self.specials,
+                ]
+                for key in mapping
+            ]
+        )
+        self.literal_words = {"one", "ones"}
+
+    def process_words(self, words: List[str]) -> Iterator[str]:
+        prefix: Optional[str] = None
+        value: Optional[Union[str, int]] = None
+        skip = False
+
+        def to_fraction(s: str):
+            try:
+                return Fraction(s)
+            except ValueError:
+                return None
+
+        def output(result: Union[str, int]):
+            nonlocal prefix, value
+            result = str(result)
+            if prefix is not None:
+                result = prefix + result
+            value = None
+            prefix = None
+            return result
+
+        if len(words) == 0:
+            return
+
+        for prev, current, next in windowed([None] + words + [None], 3):
+            if skip:
+                skip = False
+                continue
+
+            next_is_numeric = next is not None and re.match(r"^\d+(\.\d+)?$", next)
+            has_prefix = current[0] in self.prefixes
+            current_without_prefix = current[1:] if has_prefix else current
+            if re.match(r"^\d+(\.\d+)?$", current_without_prefix):
+                # arabic numbers (potentially with signs and fractions)
+                f = to_fraction(current_without_prefix)
+                assert f is not None
+                if value is not None:
+                    if isinstance(value, str) and value.endswith("."):
+                        # concatenate decimals / ip address components
+                        value = str(value) + str(current)
+                        continue
+                    else:
+                        yield output(value)
+
+                prefix = current[0] if has_prefix else prefix
+                if f.denominator == 1:
+                    value = f.numerator  # store integers as int
+                else:
+                    value = current_without_prefix
+            elif current not in self.words:
+                # non-numeric words
+                if value is not None:
+                    yield output(value)
+                yield output(current)
+            elif current in self.zeros:
+                value = str(value or "") + "0"
+            elif current in self.ones:
+                ones = self.ones[current]
+
+                if value is None:
+                    value = ones
+                elif isinstance(value, str) or prev in self.ones:
+                    if (
+                        prev in self.tens and ones < 10
+                    ):  # replace the last zero with the digit
+                        assert value[-1] == "0"
+                        value = value[:-1] + str(ones)
+                    else:
+                        value = str(value) + str(ones)
+                elif ones < 10:
+                    if value % 10 == 0:
+                        value += ones
+                    else:
+                        value = str(value) + str(ones)
+                else:  # eleven to nineteen
+                    if value % 100 == 0:
+                        value += ones
+                    else:
+                        value = str(value) + str(ones)
+            elif current in self.ones_suffixed:
+                # ordinal or cardinal; yield the number right away
+                ones, suffix = self.ones_suffixed[current]
+                if value is None:
+                    yield output(str(ones) + suffix)
+                elif isinstance(value, str) or prev in self.ones:
+                    if prev in self.tens and ones < 10:
+                        assert value[-1] == "0"
+                        yield output(value[:-1] + str(ones) + suffix)
+                    else:
+                        yield output(str(value) + str(ones) + suffix)
+                elif ones < 10:
+                    if value % 10 == 0:
+                        yield output(str(value + ones) + suffix)
+                    else:
+                        yield output(str(value) + str(ones) + suffix)
+                else:  # eleven to nineteen
+                    if value % 100 == 0:
+                        yield output(str(value + ones) + suffix)
+                    else:
+                        yield output(str(value) + str(ones) + suffix)
+                value = None
+            elif current in self.tens:
+                tens = self.tens[current]
+                if value is None:
+                    value = tens
+                elif isinstance(value, str):
+                    value = str(value) + str(tens)
+                else:
+                    if value % 100 == 0:
+                        value += tens
+                    else:
+                        value = str(value) + str(tens)
+            elif current in self.tens_suffixed:
+                # ordinal or cardinal; yield the number right away
+                tens, suffix = self.tens_suffixed[current]
+                if value is None:
+                    yield output(str(tens) + suffix)
+                elif isinstance(value, str):
+                    yield output(str(value) + str(tens) + suffix)
+                else:
+                    if value % 100 == 0:
+                        yield output(str(value + tens) + suffix)
+                    else:
+                        yield output(str(value) + str(tens) + suffix)
+            elif current in self.multipliers:
+                multiplier = self.multipliers[current]
+                if value is None:
+                    value = multiplier
+                elif isinstance(value, str) or value == 0:
+                    f = to_fraction(value)
+                    p = f * multiplier if f is not None else None
+                    if f is not None and p.denominator == 1:
+                        value = p.numerator
+                    else:
+                        yield output(value)
+                        value = multiplier
+                else:
+                    before = value // 1000 * 1000
+                    residual = value % 1000
+                    value = before + residual * multiplier
+            elif current in self.multipliers_suffixed:
+                multiplier, suffix = self.multipliers_suffixed[current]
+                if value is None:
+                    yield output(str(multiplier) + suffix)
+                elif isinstance(value, str):
+                    f = to_fraction(value)
+                    p = f * multiplier if f is not None else None
+                    if f is not None and p.denominator == 1:
+                        yield output(str(p.numerator) + suffix)
+                    else:
+                        yield output(value)
+                        yield output(str(multiplier) + suffix)
+                else:  # int
+                    before = value // 1000 * 1000
+                    residual = value % 1000
+                    value = before + residual * multiplier
+                    yield output(str(value) + suffix)
+                value = None
+            elif current in self.preceding_prefixers:
+                # apply prefix (positive, minus, etc.) if it precedes a number
+                if value is not None:
+                    yield output(value)
+
+                if next in self.words or next_is_numeric:
+                    prefix = self.preceding_prefixers[current]
+                else:
+                    yield output(current)
+            elif current in self.following_prefixers:
+                # apply prefix (dollars, cents, etc.) only after a number
+                if value is not None:
+                    prefix = self.following_prefixers[current]
+                    yield output(value)
+                else:
+                    yield output(current)
+            elif current in self.suffixers:
+                # apply suffix symbols (percent -> '%')
+                if value is not None:
+                    suffix = self.suffixers[current]
+                    if isinstance(suffix, dict):
+                        if next in suffix:
+                            yield output(str(value) + suffix[next])
+                            skip = True
+                        else:
+                            yield output(value)
+                            yield output(current)
+                    else:
+                        yield output(str(value) + suffix)
+                else:
+                    yield output(current)
+            elif current in self.specials:
+                if next not in self.words and not next_is_numeric:
+                    # apply special handling only if the next word can be numeric
+                    if value is not None:
+                        yield output(value)
+                    yield output(current)
+                elif current == "and":
+                    # ignore "and" after hundreds, thousands, etc.
+                    if prev not in self.multipliers:
+                        if value is not None:
+                            yield output(value)
+                        yield output(current)
+                elif current == "double" or current == "triple":
+                    if next in self.ones or next in self.zeros:
+                        repeats = 2 if current == "double" else 3
+                        ones = self.ones.get(next, 0)
+                        value = str(value or "") + str(ones) * repeats
+                        skip = True
+                    else:
+                        if value is not None:
+                            yield output(value)
+                        yield output(current)
+                elif current == "point":
+                    if next in self.decimals or next_is_numeric:
+                        value = str(value or "") + "."
+                else:
+                    # should all have been covered at this point
+                    raise ValueError(f"Unexpected token: {current}")
+            else:
+                # all should have been covered at this point
+                raise ValueError(f"Unexpected token: {current}")
+
+        if value is not None:
+            yield output(value)
+
+    def preprocess(self, s: str):
+        # replace "<number> and a half" with "<number> point five"
+        results = []
+
+        segments = re.split(r"\band\s+a\s+half\b", s)
+        for i, segment in enumerate(segments):
+            if len(segment.strip()) == 0:
+                continue
+            if i == len(segments) - 1:
+                results.append(segment)
+            else:
+                results.append(segment)
+                last_word = segment.rsplit(maxsplit=2)[-1]
+                if last_word in self.decimals or last_word in self.multipliers:
+                    results.append("point five")
+                else:
+                    results.append("and a half")
+
+        s = " ".join(results)
+
+        # put a space at number/letter boundary
+        s = re.sub(r"([a-z])([0-9])", r"\1 \2", s)
+        s = re.sub(r"([0-9])([a-z])", r"\1 \2", s)
+
+        # but remove spaces which could be a suffix
+        s = re.sub(r"([0-9])\s+(st|nd|rd|th|s)\b", r"\1\2", s)
+
+        return s
+
+    def postprocess(self, s: str):
+        def combine_cents(m: Match):
+            try:
+                currency = m.group(1)
+                integer = m.group(2)
+                cents = int(m.group(3))
+                return f"{currency}{integer}.{cents:02d}"
+            except ValueError:
+                return m.string
+
+        def extract_cents(m: Match):
+            try:
+                return f"¢{int(m.group(1))}"
+            except ValueError:
+                return m.string
+
+        # apply currency postprocessing; "$2 and ¢7" -> "$2.07"
+        s = re.sub(r"([€£$])([0-9]+) (?:and )?¢([0-9]{1,2})\b", combine_cents, s)
+        s = re.sub(r"[€£$]0.([0-9]{1,2})\b", extract_cents, s)
+
+        # write "one(s)" instead of "1(s)", just for the readability
+        s = re.sub(r"\b1(s?)\b", r"one\1", s)
+
+        return s
+
+    def __call__(self, s: str):
+        s = self.preprocess(s)
+        s = " ".join(word for word in self.process_words(s.split()) if word is not None)
+        s = self.postprocess(s)
+
+        return s
+
+
+class EnglishSpellingNormalizer:
+    """
+    Applies British-American spelling mappings as listed in [1].
+
+    [1] https://www.tysto.com/uk-us-spelling-list.html
+    """
+
+    def __init__(self):
+        mapping_path = os.path.join(os.path.dirname(__file__), "english.json")
+        self.mapping = json.load(open(mapping_path))
+
+    def __call__(self, s: str):
+        return " ".join(self.mapping.get(word, word) for word in s.split())
+
+
+class EnglishTextNormalizer:
+    def __init__(self):
+        self.ignore_patterns = r"\b(hmm|mm|mhm|mmm|uh|um)\b"
+        self.replacers = {
+            # common contractions
+            r"\bwon't\b": "will not",
+            r"\bcan't\b": "can not",
+            r"\blet's\b": "let us",
+            r"\bain't\b": "aint",
+            r"\by'all\b": "you all",
+            r"\bwanna\b": "want to",
+            r"\bgotta\b": "got to",
+            r"\bgonna\b": "going to",
+            r"\bi'ma\b": "i am going to",
+            r"\bimma\b": "i am going to",
+            r"\bwoulda\b": "would have",
+            r"\bcoulda\b": "could have",
+            r"\bshoulda\b": "should have",
+            r"\bma'am\b": "madam",
+            # contractions in titles/prefixes
+            r"\bmr\b": "mister ",
+            r"\bmrs\b": "missus ",
+            r"\bst\b": "saint ",
+            r"\bdr\b": "doctor ",
+            r"\bprof\b": "professor ",
+            r"\bcapt\b": "captain ",
+            r"\bgov\b": "governor ",
+            r"\bald\b": "alderman ",
+            r"\bgen\b": "general ",
+            r"\bsen\b": "senator ",
+            r"\brep\b": "representative ",
+            r"\bpres\b": "president ",
+            r"\brev\b": "reverend ",
+            r"\bhon\b": "honorable ",
+            r"\basst\b": "assistant ",
+            r"\bassoc\b": "associate ",
+            r"\blt\b": "lieutenant ",
+            r"\bcol\b": "colonel ",
+            r"\bjr\b": "junior ",
+            r"\bsr\b": "senior ",
+            r"\besq\b": "esquire ",
+            # prefect tenses, ideally it should be any past participles, but it's harder..
+            r"'d been\b": " had been",
+            r"'s been\b": " has been",
+            r"'d gone\b": " had gone",
+            r"'s gone\b": " has gone",
+            r"'d done\b": " had done",  # "'s done" is ambiguous
+            r"'s got\b": " has got",
+            # general contractions
+            r"n't\b": " not",
+            r"'re\b": " are",
+            r"'s\b": " is",
+            r"'d\b": " would",
+            r"'ll\b": " will",
+            r"'t\b": " not",
+            r"'ve\b": " have",
+            r"'m\b": " am",
+        }
+        self.standardize_numbers = EnglishNumberNormalizer()
+        self.standardize_spellings = EnglishSpellingNormalizer()
+
+    def __call__(self, s: str):
+        s = s.lower()
+
+        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
+        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
+        s = re.sub(self.ignore_patterns, "", s)
+        s = re.sub(r"\s+'", "'", s)  # when there's a space before an apostrophe
+
+        for pattern, replacement in self.replacers.items():
+            s = re.sub(pattern, replacement, s)
+
+        s = re.sub(r"(\d),(\d)", r"\1\2", s)  # remove commas between digits
+        s = re.sub(r"\.([^0-9]|$)", r" \1", s)  # remove periods not followed by numbers
+        s = remove_symbols_and_diacritics(s, keep=".%$¢€£")  # keep numeric symbols
+
+        s = self.standardize_numbers(s)
+        s = self.standardize_spellings(s)
+
+        # now remove prefix/suffix symbols that are not preceded/followed by numbers
+        s = re.sub(r"[.$¢€£]([^0-9])", r" \1", s)
+        s = re.sub(r"([^0-9])%", r"\1 ", s)
+
+        s = re.sub(r"\s+", " ", s)  # replace any successive whitespaces with a space
+
+        return s
--- a/tests/en-0-ref.txt
+++ b/tests/en-0-ref.txt
@@ -0,0 +1 @@
+ My fellow Americans, this day has brought terrible news and great sadness to our country. At 9 o'clock this morning, Mission Control in Houston lost contact with our space shuttle, Columbia. A short time later, debris was seen falling from the skies above Texas. The Colombians lost. There are no survivors. On board was a crew of seven. Colonel Rick Husband, Lieutenant Colonel Michael Anderson, Commander Laurel Clark, Captain David Brown, Commander William McCool, Dr. Kultna Shavla, and Ilan Ramon, a colonel in the Israeli Air Force. These men and women assumed great risk in the service to all humanity. In an age when spaceflight has come to seem almost routine, it is easy to overlook the dangers of travel by rocket and the difficulties of navigating the fierce outer atmosphere of the Earth. These astronauts knew the dangers, and they faced them willingly, knowing they had a high and noble purpose in life. Because of their courage and daring and idealism, we will miss them all the more. All Americans today are thinking as well of the families of these men and women who have been given this sudden shock and grief. You're not alone. Our entire nation grieves with you. And those you love will always have the respect and gratitude of this country. The cause in which they died will continue. Mankind is led into the darkness beyond our world by the inspiration of discovery and the longing to understand. Our journey into space will go on. In the skies today, we saw destruction and tragedy. Yet farther than we can see, there is comfort and hope. In the words of the prophet Isaiah, "Lift your eyes and look to the heavens. Who created all these? He who brings out the starry hosts one by one and calls them each by name." Because of His great power and mighty strength, not one of them is missing. The same Creator who names the stars also knows the names of the seven souls we mourn today. The crew of the shuttle Columbia did not return safely to Earth, yet we can pray that all are safely home. May God bless the grieving families. And may God continue to bless America. [Silence]
--- a/tests/en-1-ref.txt
+++ b/tests/en-1-ref.txt
@@ -0,0 +1 @@
+ Henry F. Phillips from Wikipedia, the free encyclopedia at en.wikipedia.org. Henry F. Phillips from Wikipedia, the free encyclopedia. Henry F. Phillips 1890-1958, a U.S. businessman from Portland, Oregon, has the honor of having the Phillips head screw and screwdriver named after him. The importance of the cross head screw design lies in its self-centering property, useful on automated production lines that use powered screwdrivers. Phillips' major contribution was in driving the cross head concept forward to the point where it was adopted by screw makers and automobile companies. Although he received patents for the design in 1936, U.S. Patent #2,046,343, U.S. Patents #2,046,837 to #2,046,840, it was so widely copied that by 1949 Phillips lost his patent. The American Screw Company was responsible for devising a means of manufacturing the screw, and successfully patented and licensed their method. Other screw makers of the 1930s dismissed the Phillips concept since it calls for a relatively complex recessed socket shape in the head of the screw, as distinct from the simple milled slot of a slotted type screw. The Phillips Screw Company and the American Screw Company went on to devise the Pawsadrive screw, which differs from the Phillips in that it is designed to accommodate greater torque than the Phillips. An image accompanied this article, captioned "Phillips Screw Head." The following is an info box which accompanies this article. Info box, part of the series on screw drive types. Slotted, commonly erroneously flat head. Phillips, cross head. Pawsadrive, super drive. Torques. Hex, Allen. Robertson. Tri-wing. Torx set. Spanner head. Triple square, XZN. Others, poly drive, spline drive, double hex. Many images accompanied this info box. This page was last modified on the 9th of April, 2008, at 1704. All text is available under the terms of the GNU Free Documentation License. See copyrights for details. Wikipedia is a registered trademark of the Wikimedia Foundation Incorporated, a U.S. registered 501(c)(3) tax-deductible nonprofit charity. This sound file and all text in the article are licensed under the GNU Free Documentation License, available at www.gnu.org/copyleft/fdl.html.
--- a/tests/en-2-ref.txt
+++ b/tests/en-2-ref.txt
@@ -0,0 +1 @@
+ This is the Micro Machine Man presenting the most midget miniature motorcade of Micro Machines. Each one has dramatic details, terrific trim, precision paint jobs, plus incredible Micro Machine Pocket Playsets. There's a police station, fire station, restaurant, service station, and more. Perfect pocket portables to take anyplace. And there are many miniature playsets to play with, and each one comes with its own special edition Micro Machine vehicle and fun, fantastic features that miraculously move. Raise the boat lift at the airport marina, man the gun turret at the army base, clean your car at the car wash, raise the toll bridge. And these playsets fit together to form a Micro Machine world. Micro Machine Pocket Playsets, so tremendously tiny, so perfectly precise, so dazzlingly detailed, you'll want to pocket them all. Micro Machines are Micro Machine Pocket Playsets sold separately from Galoob. The smaller they are, the better they are.
--- a/tests/es-0-ref.txt
+++ b/tests/es-0-ref.txt
@@ -0,0 +1 @@
+ Hola, como están todos? Mi nombre es Julián Virrueta Mendoza y en este podcast les vengo a hablar sobre la contaminación del agua. Bueno, empezaré por decir que el ser humano no está midiendo las consecuencias de sus actos. No hay duda que uno de los mayores problemas a los que se enfrentan muchas poblaciones actualmente es la contaminación del agua. Principalmente porque como bien sabemos el agua prácticamente es fundamental para la vida, por lo que la contaminación puede ser algo muy negativo para el desarrollo tanto económico como social de los pueblos o de las poblaciones próximas en ese lugar contaminado. Los comienzos de la contaminación, como lo definen muchos expertos en la materia, la contaminación del agua es causada por las actividades humanas. Es un fenómeno ambiental de importancia, el cual se comienza a producir desde los primeros intentos de industrialización para transformarse luego en un problema tan habitual como generalizado. Generalmente la contaminación del agua se produce a través de la introducción directa o indirecta en los acuíferos o caos de agua, ríos, mares, lagos, océanos, etc. o de diversas sustancias que pueden ser consideradas como contaminantes. Pero existen dos formas principales de contaminación del agua. Una de ellas tiene que ver con la contaminación natural del agua que se corresponde con el ciclo natural de esta durante el que puede entrar en contacto con ciertos constituyentes contaminantes como sustancias minerales y orgánicas disueltas o en suspensión que se vierten en la corteza terrestre, la atmósfera y en las aguas. Pero todo esto se puede contradecir si el ser humano comía sus consecuencias, si no tirara basura a los lagos, a los ríos, no tirara botes de aceite, no contaminara. Bueno amigos, yo los invito a que no contaminen el agua y que sepan cuidar la naturaleza. Los saluda su buen amigo y compañero Julián Virreta. Nos vemos. ¡Claro!
--- a/tests/librispeech-parakeet/.gitignore
+++ b/tests/librispeech-parakeet/.gitignore
@@ -0,0 +1,6 @@
+__pycache__
+*.tar.gz
+*.txt
+eval.conf
+venv
+LibriSpeech
--- a/tests/librispeech-parakeet/Makefile
+++ b/tests/librispeech-parakeet/Makefile
@@ -0,0 +1,15 @@
+TAR_URL = https://www.openslr.org/resources/12/test-clean.tar.gz
+
+all: eval
+
+eval:
+	$(MAKE) -f eval.mk
+
+clean:
+	$(MAKE) -f eval.mk clean
+
+get-audio:
+	wget -c $(TAR_URL)
+	tar -xf test-clean.tar.gz
+
+.PHONY: all eval clean setup-venv clean-venv get-audio
--- a/tests/librispeech-parakeet/README.md
+++ b/tests/librispeech-parakeet/README.md
@@ -0,0 +1,57 @@
+# parakeet.cpp/tests/librispeech
+
+[LibriSpeech](https://www.openslr.org/12) is a standard dataset for
+training and evaluating automatic speech recognition systems.
+
+This directory contains a set of tools to evaluate the recognition
+performance of parakeet.cpp on LibriSpeech corpus.
+
+## Quick Start
+
+1. (Pre-requirement) Compile `parakeet-cli` and prepare the Parakeet
+   model in `ggml` format.
+
+   ```
+   $ # Execute the commands below in the project root dir.
+   $ cmake -B build
+   $ cmake --build build --config Release
+   ```
+
+2. Download the audio files from LibriSpeech project.
+
+   ```
+   $ make get-audio
+   ```
+
+3. Set up the environment to compute WER score.
+
+   ```
+   $ pip install -r requirements.txt
+   ```
+
+   For example, if you use `virtualenv`, you can set up it as follows:
+
+   ```
+   $ python3 -m venv venv
+   $ . venv/bin/activate
+   $ pip install -r requirements.txt
+   ```
+
+4. Run the benchmark test.
+
+   ```
+   $ make
+   ```
+
+## How-to guides
+
+### How to change the inference parameters
+
+Create `eval.conf` and override variables.
+
+```
+PARAKEET_MODEL = parakeet-tdt-0.6b-v3
+PARAKEET_FLAGS = --no-prints --threads 8 --language en --output-txt
+```
+
+Check out `eval.mk` for more details.
--- a/tests/librispeech-parakeet/eval.mk
+++ b/tests/librispeech-parakeet/eval.mk
@@ -0,0 +1,39 @@
+PYTHON = python
+
+PARAKEET_PREFIX = ../../
+PARAKEET_MODEL = parakeet-tdt-0.6b-v3
+
+PARAKEET_CLI = $(PARAKEET_PREFIX)build/bin/parakeet-cli
+PARAKEET_FLAGS = --no-prints --output-txt
+
+# You can create eval.conf to override the PARAKEET_* variables
+# defined above.
+-include eval.conf
+
+# This follows the file structure of the LibriSpeech project.
+AUDIO_SRCS = $(sort $(wildcard LibriSpeech/*/*/*/*.flac))
+TRANS_TXTS = $(addsuffix .txt, $(AUDIO_SRCS))
+
+# We output the evaluation result to this file.
+DONE = $(PARAKEET_MODEL).txt
+
+all: $(DONE)
+
+$(DONE): $(TRANS_TXTS)
+	$(PYTHON) eval.py > $@.tmp
+	mv $@.tmp $@
+
+# Note: This task writes to a temporary file first to
+# create the target file atomically.
+%.flac.txt: %.flac
+	$(PARAKEET_CLI) $(PARAKEET_FLAGS) --model $(PARAKEET_PREFIX)models/ggml-$(PARAKEET_MODEL).bin --file $^ --output-file $^.tmp
+	mv $^.tmp.txt $^.txt
+
+archive:
+	tar -czf $(PARAKEET_MODEL).tar.gz --exclude="*.flac" LibriSpeech $(DONE)
+
+clean:
+	@rm -f $(TRANS_TXTS)
+	@rm -f $(DONE)
+
+.PHONY: all clean
--- a/tests/librispeech-parakeet/eval.py
+++ b/tests/librispeech-parakeet/eval.py
@@ -0,0 +1,47 @@
+import os
+import glob
+import jiwer
+from normalizers import EnglishTextNormalizer
+
+def get_reference():
+    ref = {}
+    for path in glob.glob('LibriSpeech/*/*/*/*.trans.txt'):
+        with open(path) as fp:
+            for line in fp:
+                code, text = line.strip().split(" ", maxsplit=1)
+                ref [code] = text
+    return ref
+
+def get_hypothesis():
+    hyp = {}
+    for path in glob.glob('LibriSpeech/*/*/*/*.flac.txt'):
+        with open(path) as fp:
+            text = fp.read().strip()
+        code = os.path.basename(path).replace('.flac.txt', '')
+        hyp[code] = text
+    return hyp
+
+def get_codes():
+    codes = []
+    for path in glob.glob('LibriSpeech/*/*/*/*.flac'):
+        codes.append(os.path.basename(path).replace('.flac', ''))
+    return sorted(codes)
+
+def main():
+    normalizer = EnglishTextNormalizer()
+
+    ref_orig = get_reference()
+    hyp_orig = get_hypothesis()
+
+    ref_clean = []
+    hyp_clean = []
+
+    for code in get_codes():
+        ref_clean.append(normalizer(ref_orig[code]))
+        hyp_clean.append(normalizer(hyp_orig[code]))
+
+    wer = jiwer.wer(ref_clean, hyp_clean)
+    print(f"WER: {wer * 100:.2f}%")
+
+if __name__ == '__main__':
+    main()
--- a/tests/librispeech-parakeet/normalizers/LICENSE
+++ b/tests/librispeech-parakeet/normalizers/LICENSE
@@ -0,0 +1,25 @@
+Code in this directory is adapted from OpenAI Whisper project
+(https://github.com/openai/whisper) and carries the following
+copyright and license.
+
+    MIT License
+
+    Copyright (c) 2022 OpenAI
+
+    Permission is hereby granted, free of charge, to any person obtaining a copy
+    of this software and associated documentation files (the "Software"), to deal
+    in the Software without restriction, including without limitation the rights
+    to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+    copies of the Software, and to permit persons to whom the Software is
+    furnished to do so, subject to the following conditions:
+
+    The above copyright notice and this permission notice shall be included in all
+    copies or substantial portions of the Software.
+
+    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+    OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+    SOFTWARE.
--- a/tests/librispeech-parakeet/normalizers/init.py
+++ b/tests/librispeech-parakeet/normalizers/init.py
@@ -0,0 +1,2 @@
+from .basic import BasicTextNormalizer as BasicTextNormalizer
+from .english import EnglishTextNormalizer as EnglishTextNormalizer
--- a/tests/librispeech-parakeet/normalizers/basic.py
+++ b/tests/librispeech-parakeet/normalizers/basic.py
@@ -0,0 +1,80 @@
+import re
+import unicodedata
+
+import regex
+
+# non-ASCII letters that are not separated by "NFKD" normalization
+ADDITIONAL_DIACRITICS = {
+    "œ": "oe",
+    "Œ": "OE",
+    "ø": "o",
+    "Ø": "O",
+    "æ": "ae",
+    "Æ": "AE",
+    "ß": "ss",
+    "ẞ": "SS",
+    "đ": "d",
+    "Đ": "D",
+    "ð": "d",
+    "Ð": "D",
+    "þ": "th",
+    "Þ": "th",
+    "ł": "l",
+    "Ł": "L",
+}
+
+
+def remove_symbols_and_diacritics(s: str, keep=""):
+    """
+    Replace any other markers, symbols, and punctuations with a space,
+    and drop any diacritics (category 'Mn' and some manual mappings)
+    """
+    return "".join(
+        (
+            c
+            if c in keep
+            else (
+                ADDITIONAL_DIACRITICS[c]
+                if c in ADDITIONAL_DIACRITICS
+                else (
+                    ""
+                    if unicodedata.category(c) == "Mn"
+                    else " " if unicodedata.category(c)[0] in "MSP" else c
+                )
+            )
+        )
+        for c in unicodedata.normalize("NFKD", s)
+    )
+
+
+def remove_symbols(s: str):
+    """
+    Replace any other markers, symbols, punctuations with a space, keeping diacritics
+    """
+    return "".join(
+        " " if unicodedata.category(c)[0] in "MSP" else c
+        for c in unicodedata.normalize("NFKC", s)
+    )
+
+
+class BasicTextNormalizer:
+    def __init__(self, remove_diacritics: bool = False, split_letters: bool = False):
+        self.clean = (
+            remove_symbols_and_diacritics if remove_diacritics else remove_symbols
+        )
+        self.split_letters = split_letters
+
+    def __call__(self, s: str):
+        s = s.lower()
+        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
+        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
+        s = self.clean(s).lower()
+
+        if self.split_letters:
+            s = " ".join(regex.findall(r"\X", s, regex.U))
+
+        s = re.sub(
+            r"\s+", " ", s
+        )  # replace any successive whitespace characters with a space
+
+        return s
--- a/tests/librispeech-parakeet/normalizers/english.json
+++ b/tests/librispeech-parakeet/normalizers/english.json
--- a/tests/librispeech-parakeet/normalizers/english.py
+++ b/tests/librispeech-parakeet/normalizers/english.py
@@ -0,0 +1,550 @@
+import json
+import os
+import re
+from fractions import Fraction
+from typing import Iterator, List, Match, Optional, Union
+
+from more_itertools import windowed
+
+from .basic import remove_symbols_and_diacritics
+
+
+class EnglishNumberNormalizer:
+    """
+    Convert any spelled-out numbers into arabic numbers, while handling:
+
+    - remove any commas
+    - keep the suffixes such as: `1960s`, `274th`, `32nd`, etc.
+    - spell out currency symbols after the number. e.g. `$20 million` -> `20000000 dollars`
+    - spell out `one` and `ones`
+    - interpret successive single-digit numbers as nominal: `one oh one` -> `101`
+    """
+
+    def __init__(self):
+        super().__init__()
+
+        self.zeros = {"o", "oh", "zero"}
+        self.ones = {
+            name: i
+            for i, name in enumerate(
+                [
+                    "one",
+                    "two",
+                    "three",
+                    "four",
+                    "five",
+                    "six",
+                    "seven",
+                    "eight",
+                    "nine",
+                    "ten",
+                    "eleven",
+                    "twelve",
+                    "thirteen",
+                    "fourteen",
+                    "fifteen",
+                    "sixteen",
+                    "seventeen",
+                    "eighteen",
+                    "nineteen",
+                ],
+                start=1,
+            )
+        }
+        self.ones_plural = {
+            "sixes" if name == "six" else name + "s": (value, "s")
+            for name, value in self.ones.items()
+        }
+        self.ones_ordinal = {
+            "zeroth": (0, "th"),
+            "first": (1, "st"),
+            "second": (2, "nd"),
+            "third": (3, "rd"),
+            "fifth": (5, "th"),
+            "twelfth": (12, "th"),
+            **{
+                name + ("h" if name.endswith("t") else "th"): (value, "th")
+                for name, value in self.ones.items()
+                if value > 3 and value != 5 and value != 12
+            },
+        }
+        self.ones_suffixed = {**self.ones_plural, **self.ones_ordinal}
+
+        self.tens = {
+            "twenty": 20,
+            "thirty": 30,
+            "forty": 40,
+            "fifty": 50,
+            "sixty": 60,
+            "seventy": 70,
+            "eighty": 80,
+            "ninety": 90,
+        }
+        self.tens_plural = {
+            name.replace("y", "ies"): (value, "s") for name, value in self.tens.items()
+        }
+        self.tens_ordinal = {
+            name.replace("y", "ieth"): (value, "th")
+            for name, value in self.tens.items()
+        }
+        self.tens_suffixed = {**self.tens_plural, **self.tens_ordinal}
+
+        self.multipliers = {
+            "hundred": 100,
+            "thousand": 1_000,
+            "million": 1_000_000,
+            "billion": 1_000_000_000,
+            "trillion": 1_000_000_000_000,
+            "quadrillion": 1_000_000_000_000_000,
+            "quintillion": 1_000_000_000_000_000_000,
+            "sextillion": 1_000_000_000_000_000_000_000,
+            "septillion": 1_000_000_000_000_000_000_000_000,
+            "octillion": 1_000_000_000_000_000_000_000_000_000,
+            "nonillion": 1_000_000_000_000_000_000_000_000_000_000,
+            "decillion": 1_000_000_000_000_000_000_000_000_000_000_000,
+        }
+        self.multipliers_plural = {
+            name + "s": (value, "s") for name, value in self.multipliers.items()
+        }
+        self.multipliers_ordinal = {
+            name + "th": (value, "th") for name, value in self.multipliers.items()
+        }
+        self.multipliers_suffixed = {
+            **self.multipliers_plural,
+            **self.multipliers_ordinal,
+        }
+        self.decimals = {*self.ones, *self.tens, *self.zeros}
+
+        self.preceding_prefixers = {
+            "minus": "-",
+            "negative": "-",
+            "plus": "+",
+            "positive": "+",
+        }
+        self.following_prefixers = {
+            "pound": "£",
+            "pounds": "£",
+            "euro": "€",
+            "euros": "€",
+            "dollar": "$",
+            "dollars": "$",
+            "cent": "¢",
+            "cents": "¢",
+        }
+        self.prefixes = set(
+            list(self.preceding_prefixers.values())
+            + list(self.following_prefixers.values())
+        )
+        self.suffixers = {
+            "per": {"cent": "%"},
+            "percent": "%",
+        }
+        self.specials = {"and", "double", "triple", "point"}
+
+        self.words = set(
+            [
+                key
+                for mapping in [
+                    self.zeros,
+                    self.ones,
+                    self.ones_suffixed,
+                    self.tens,
+                    self.tens_suffixed,
+                    self.multipliers,
+                    self.multipliers_suffixed,
+                    self.preceding_prefixers,
+                    self.following_prefixers,
+                    self.suffixers,
+                    self.specials,
+                ]
+                for key in mapping
+            ]
+        )
+        self.literal_words = {"one", "ones"}
+
+    def process_words(self, words: List[str]) -> Iterator[str]:
+        prefix: Optional[str] = None
+        value: Optional[Union[str, int]] = None
+        skip = False
+
+        def to_fraction(s: str):
+            try:
+                return Fraction(s)
+            except ValueError:
+                return None
+
+        def output(result: Union[str, int]):
+            nonlocal prefix, value
+            result = str(result)
+            if prefix is not None:
+                result = prefix + result
+            value = None
+            prefix = None
+            return result
+
+        if len(words) == 0:
+            return
+
+        for prev, current, next in windowed([None] + words + [None], 3):
+            if skip:
+                skip = False
+                continue
+
+            next_is_numeric = next is not None and re.match(r"^\d+(\.\d+)?$", next)
+            has_prefix = current[0] in self.prefixes
+            current_without_prefix = current[1:] if has_prefix else current
+            if re.match(r"^\d+(\.\d+)?$", current_without_prefix):
+                # arabic numbers (potentially with signs and fractions)
+                f = to_fraction(current_without_prefix)
+                assert f is not None
+                if value is not None:
+                    if isinstance(value, str) and value.endswith("."):
+                        # concatenate decimals / ip address components
+                        value = str(value) + str(current)
+                        continue
+                    else:
+                        yield output(value)
+
+                prefix = current[0] if has_prefix else prefix
+                if f.denominator == 1:
+                    value = f.numerator  # store integers as int
+                else:
+                    value = current_without_prefix
+            elif current not in self.words:
+                # non-numeric words
+                if value is not None:
+                    yield output(value)
+                yield output(current)
+            elif current in self.zeros:
+                value = str(value or "") + "0"
+            elif current in self.ones:
+                ones = self.ones[current]
+
+                if value is None:
+                    value = ones
+                elif isinstance(value, str) or prev in self.ones:
+                    if (
+                        prev in self.tens and ones < 10
+                    ):  # replace the last zero with the digit
+                        assert value[-1] == "0"
+                        value = value[:-1] + str(ones)
+                    else:
+                        value = str(value) + str(ones)
+                elif ones < 10:
+                    if value % 10 == 0:
+                        value += ones
+                    else:
+                        value = str(value) + str(ones)
+                else:  # eleven to nineteen
+                    if value % 100 == 0:
+                        value += ones
+                    else:
+                        value = str(value) + str(ones)
+            elif current in self.ones_suffixed:
+                # ordinal or cardinal; yield the number right away
+                ones, suffix = self.ones_suffixed[current]
+                if value is None:
+                    yield output(str(ones) + suffix)
+                elif isinstance(value, str) or prev in self.ones:
+                    if prev in self.tens and ones < 10:
+                        assert value[-1] == "0"
+                        yield output(value[:-1] + str(ones) + suffix)
+                    else:
+                        yield output(str(value) + str(ones) + suffix)
+                elif ones < 10:
+                    if value % 10 == 0:
+                        yield output(str(value + ones) + suffix)
+                    else:
+                        yield output(str(value) + str(ones) + suffix)
+                else:  # eleven to nineteen
+                    if value % 100 == 0:
+                        yield output(str(value + ones) + suffix)
+                    else:
+                        yield output(str(value) + str(ones) + suffix)
+                value = None
+            elif current in self.tens:
+                tens = self.tens[current]
+                if value is None:
+                    value = tens
+                elif isinstance(value, str):
+                    value = str(value) + str(tens)
+                else:
+                    if value % 100 == 0:
+                        value += tens
+                    else:
+                        value = str(value) + str(tens)
+            elif current in self.tens_suffixed:
+                # ordinal or cardinal; yield the number right away
+                tens, suffix = self.tens_suffixed[current]
+                if value is None:
+                    yield output(str(tens) + suffix)
+                elif isinstance(value, str):
+                    yield output(str(value) + str(tens) + suffix)
+                else:
+                    if value % 100 == 0:
+                        yield output(str(value + tens) + suffix)
+                    else:
+                        yield output(str(value) + str(tens) + suffix)
+            elif current in self.multipliers:
+                multiplier = self.multipliers[current]
+                if value is None:
+                    value = multiplier
+                elif isinstance(value, str) or value == 0:
+                    f = to_fraction(value)
+                    p = f * multiplier if f is not None else None
+                    if f is not None and p.denominator == 1:
+                        value = p.numerator
+                    else:
+                        yield output(value)
+                        value = multiplier
+                else:
+                    before = value // 1000 * 1000
+                    residual = value % 1000
+                    value = before + residual * multiplier
+            elif current in self.multipliers_suffixed:
+                multiplier, suffix = self.multipliers_suffixed[current]
+                if value is None:
+                    yield output(str(multiplier) + suffix)
+                elif isinstance(value, str):
+                    f = to_fraction(value)
+                    p = f * multiplier if f is not None else None
+                    if f is not None and p.denominator == 1:
+                        yield output(str(p.numerator) + suffix)
+                    else:
+                        yield output(value)
+                        yield output(str(multiplier) + suffix)
+                else:  # int
+                    before = value // 1000 * 1000
+                    residual = value % 1000
+                    value = before + residual * multiplier
+                    yield output(str(value) + suffix)
+                value = None
+            elif current in self.preceding_prefixers:
+                # apply prefix (positive, minus, etc.) if it precedes a number
+                if value is not None:
+                    yield output(value)
+
+                if next in self.words or next_is_numeric:
+                    prefix = self.preceding_prefixers[current]
+                else:
+                    yield output(current)
+            elif current in self.following_prefixers:
+                # apply prefix (dollars, cents, etc.) only after a number
+                if value is not None:
+                    prefix = self.following_prefixers[current]
+                    yield output(value)
+                else:
+                    yield output(current)
+            elif current in self.suffixers:
+                # apply suffix symbols (percent -> '%')
+                if value is not None:
+                    suffix = self.suffixers[current]
+                    if isinstance(suffix, dict):
+                        if next in suffix:
+                            yield output(str(value) + suffix[next])
+                            skip = True
+                        else:
+                            yield output(value)
+                            yield output(current)
+                    else:
+                        yield output(str(value) + suffix)
+                else:
+                    yield output(current)
+            elif current in self.specials:
+                if next not in self.words and not next_is_numeric:
+                    # apply special handling only if the next word can be numeric
+                    if value is not None:
+                        yield output(value)
+                    yield output(current)
+                elif current == "and":
+                    # ignore "and" after hundreds, thousands, etc.
+                    if prev not in self.multipliers:
+                        if value is not None:
+                            yield output(value)
+                        yield output(current)
+                elif current == "double" or current == "triple":
+                    if next in self.ones or next in self.zeros:
+                        repeats = 2 if current == "double" else 3
+                        ones = self.ones.get(next, 0)
+                        value = str(value or "") + str(ones) * repeats
+                        skip = True
+                    else:
+                        if value is not None:
+                            yield output(value)
+                        yield output(current)
+                elif current == "point":
+                    if next in self.decimals or next_is_numeric:
+                        value = str(value or "") + "."
+                else:
+                    # should all have been covered at this point
+                    raise ValueError(f"Unexpected token: {current}")
+            else:
+                # all should have been covered at this point
+                raise ValueError(f"Unexpected token: {current}")
+
+        if value is not None:
+            yield output(value)
+
+    def preprocess(self, s: str):
+        # replace "<number> and a half" with "<number> point five"
+        results = []
+
+        segments = re.split(r"\band\s+a\s+half\b", s)
+        for i, segment in enumerate(segments):
+            if len(segment.strip()) == 0:
+                continue
+            if i == len(segments) - 1:
+                results.append(segment)
+            else:
+                results.append(segment)
+                last_word = segment.rsplit(maxsplit=2)[-1]
+                if last_word in self.decimals or last_word in self.multipliers:
+                    results.append("point five")
+                else:
+                    results.append("and a half")
+
+        s = " ".join(results)
+
+        # put a space at number/letter boundary
+        s = re.sub(r"([a-z])([0-9])", r"\1 \2", s)
+        s = re.sub(r"([0-9])([a-z])", r"\1 \2", s)
+
+        # but remove spaces which could be a suffix
+        s = re.sub(r"([0-9])\s+(st|nd|rd|th|s)\b", r"\1\2", s)
+
+        return s
+
+    def postprocess(self, s: str):
+        def combine_cents(m: Match):
+            try:
+                currency = m.group(1)
+                integer = m.group(2)
+                cents = int(m.group(3))
+                return f"{currency}{integer}.{cents:02d}"
+            except ValueError:
+                return m.string
+
+        def extract_cents(m: Match):
+            try:
+                return f"¢{int(m.group(1))}"
+            except ValueError:
+                return m.string
+
+        # apply currency postprocessing; "$2 and ¢7" -> "$2.07"
+        s = re.sub(r"([€£$])([0-9]+) (?:and )?¢([0-9]{1,2})\b", combine_cents, s)
+        s = re.sub(r"[€£$]0.([0-9]{1,2})\b", extract_cents, s)
+
+        # write "one(s)" instead of "1(s)", just for the readability
+        s = re.sub(r"\b1(s?)\b", r"one\1", s)
+
+        return s
+
+    def __call__(self, s: str):
+        s = self.preprocess(s)
+        s = " ".join(word for word in self.process_words(s.split()) if word is not None)
+        s = self.postprocess(s)
+
+        return s
+
+
+class EnglishSpellingNormalizer:
+    """
+    Applies British-American spelling mappings as listed in [1].
+
+    [1] https://www.tysto.com/uk-us-spelling-list.html
+    """
+
+    def __init__(self):
+        mapping_path = os.path.join(os.path.dirname(__file__), "english.json")
+        self.mapping = json.load(open(mapping_path))
+
+    def __call__(self, s: str):
+        return " ".join(self.mapping.get(word, word) for word in s.split())
+
+
+class EnglishTextNormalizer:
+    def __init__(self):
+        self.ignore_patterns = r"\b(hmm|mm|mhm|mmm|uh|um)\b"
+        self.replacers = {
+            # common contractions
+            r"\bwon't\b": "will not",
+            r"\bcan't\b": "can not",
+            r"\blet's\b": "let us",
+            r"\bain't\b": "aint",
+            r"\by'all\b": "you all",
+            r"\bwanna\b": "want to",
+            r"\bgotta\b": "got to",
+            r"\bgonna\b": "going to",
+            r"\bi'ma\b": "i am going to",
+            r"\bimma\b": "i am going to",
+            r"\bwoulda\b": "would have",
+            r"\bcoulda\b": "could have",
+            r"\bshoulda\b": "should have",
+            r"\bma'am\b": "madam",
+            # contractions in titles/prefixes
+            r"\bmr\b": "mister ",
+            r"\bmrs\b": "missus ",
+            r"\bst\b": "saint ",
+            r"\bdr\b": "doctor ",
+            r"\bprof\b": "professor ",
+            r"\bcapt\b": "captain ",
+            r"\bgov\b": "governor ",
+            r"\bald\b": "alderman ",
+            r"\bgen\b": "general ",
+            r"\bsen\b": "senator ",
+            r"\brep\b": "representative ",
+            r"\bpres\b": "president ",
+            r"\brev\b": "reverend ",
+            r"\bhon\b": "honorable ",
+            r"\basst\b": "assistant ",
+            r"\bassoc\b": "associate ",
+            r"\blt\b": "lieutenant ",
+            r"\bcol\b": "colonel ",
+            r"\bjr\b": "junior ",
+            r"\bsr\b": "senior ",
+            r"\besq\b": "esquire ",
+            # prefect tenses, ideally it should be any past participles, but it's harder..
+            r"'d been\b": " had been",
+            r"'s been\b": " has been",
+            r"'d gone\b": " had gone",
+            r"'s gone\b": " has gone",
+            r"'d done\b": " had done",  # "'s done" is ambiguous
+            r"'s got\b": " has got",
+            # general contractions
+            r"n't\b": " not",
+            r"'re\b": " are",
+            r"'s\b": " is",
+            r"'d\b": " would",
+            r"'ll\b": " will",
+            r"'t\b": " not",
+            r"'ve\b": " have",
+            r"'m\b": " am",
+        }
+        self.standardize_numbers = EnglishNumberNormalizer()
+        self.standardize_spellings = EnglishSpellingNormalizer()
+
+    def __call__(self, s: str):
+        s = s.lower()
+
+        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
+        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
+        s = re.sub(self.ignore_patterns, "", s)
+        s = re.sub(r"\s+'", "'", s)  # when there's a space before an apostrophe
+
+        for pattern, replacement in self.replacers.items():
+            s = re.sub(pattern, replacement, s)
+
+        s = re.sub(r"(\d),(\d)", r"\1\2", s)  # remove commas between digits
+        s = re.sub(r"\.([^0-9]|$)", r" \1", s)  # remove periods not followed by numbers
+        s = remove_symbols_and_diacritics(s, keep=".%$¢€£")  # keep numeric symbols
+
+        s = self.standardize_numbers(s)
+        s = self.standardize_spellings(s)
+
+        # now remove prefix/suffix symbols that are not preceded/followed by numbers
+        s = re.sub(r"[.$¢€£]([^0-9])", r" \1", s)
+        s = re.sub(r"([^0-9])%", r"\1 ", s)
+
+        s = re.sub(r"\s+", " ", s)  # replace any successive whitespaces with a space
+
+        return s
--- a/tests/librispeech/.gitignore
+++ b/tests/librispeech/.gitignore
@@ -0,0 +1,6 @@
+__pycache__
+*.tar.gz
+*.txt
+eval.conf
+venv
+LibriSpeech
--- a/tests/librispeech/Makefile
+++ b/tests/librispeech/Makefile
@@ -0,0 +1,15 @@
+TAR_URL = https://www.openslr.org/resources/12/test-clean.tar.gz
+
+all: eval
+
+eval:
+	$(MAKE) -f eval.mk
+
+clean:
+	$(MAKE) -f eval.mk clean
+
+get-audio:
+	wget -c $(TAR_URL)
+	tar -xf test-clean.tar.gz
+
+.PHONY: all eval clean setup-venv clean-venv get-audio
--- a/tests/librispeech/README.md
+++ b/tests/librispeech/README.md
@@ -0,0 +1,60 @@
+# whisper.cpp/tests/librispeech
+
+[LibriSpeech](https://www.openslr.org/12) is a standard dataset for
+training and evaluating automatic speech recognition systems.
+
+This directory contains a set of tools to evaluate the recognition
+performance of whisper.cpp on LibriSpeech corpus.
+
+## Quick Start
+
+1. (Pre-requirement) Compile `whisper-cli` and prepare the Whisper
+   model in `ggml` format.
+
+   ```
+   $ # Execute the commands below in the project root dir.
+   $ cmake -B build
+   $ cmake --build build --config Release
+   $ ./models/download-ggml-model.sh tiny
+   ```
+
+   Consult [whisper.cpp/README.md](../../README.md) for more details.
+
+2. Download the audio files from LibriSpeech project.
+
+   ```
+   $ make get-audio
+   ```
+
+3. Set up the environment to compute WER score.
+
+   ```
+   $ pip install -r requirements.txt
+   ```
+
+   For example, if you use `virtualenv`, you can set up it as follows:
+
+   ```
+   $ python3 -m venv venv
+   $ . venv/bin/activate
+   $ pip install -r requirements.txt
+   ```
+
+4. Run the benchmark test.
+
+   ```
+   $ make
+   ```
+
+## How-to guides
+
+### How to change the inference parameters
+
+Create `eval.conf` and override variables.
+
+```
+WHISPER_MODEL = large-v3-turbo
+WHISPER_FLAGS = --no-prints --threads 8 --language en --output-txt
+```
+
+Check out `eval.mk` for more details.
--- a/tests/librispeech/eval.mk
+++ b/tests/librispeech/eval.mk
@@ -0,0 +1,39 @@
+PYTHON = python
+
+WHISPER_PREFIX = ../../
+WHISPER_MODEL = tiny
+
+WHISPER_CLI = $(WHISPER_PREFIX)build/bin/whisper-cli
+WHISPER_FLAGS = --no-prints --language en --output-txt
+
+# You can create eval.conf to override the WHISPER_* variables
+# defined above.
+-include eval.conf
+
+# This follows the file structure of the LibriSpeech project.
+AUDIO_SRCS = $(sort $(wildcard LibriSpeech/*/*/*/*.flac))
+TRANS_TXTS = $(addsuffix .txt, $(AUDIO_SRCS))
+
+# We output the evaluation result to this file.
+DONE = $(WHISPER_MODEL).txt
+
+all: $(DONE)
+
+$(DONE): $(TRANS_TXTS)
+	$(PYTHON) eval.py > $@.tmp
+	mv $@.tmp $@
+
+# Note: This task writes to a temporary file first to
+# create the target file atomically.
+%.flac.txt: %.flac
+	$(WHISPER_CLI) $(WHISPER_FLAGS) --model $(WHISPER_PREFIX)models/ggml-$(WHISPER_MODEL).bin --file $^ --output-file $^.tmp
+	mv $^.tmp.txt $^.txt
+
+archive:
+	tar -czf $(WHISPER_MODEL).tar.gz --exclude="*.flac" LibriSpeech $(DONE)
+
+clean:
+	@rm -f $(TRANS_TXTS)
+	@rm -f $(DONE)
+
+.PHONY: all clean
--- a/tests/librispeech/eval.py
+++ b/tests/librispeech/eval.py
@@ -0,0 +1,47 @@
+import os
+import glob
+import jiwer
+from normalizers import EnglishTextNormalizer
+
+def get_reference():
+    ref = {}
+    for path in glob.glob('LibriSpeech/*/*/*/*.trans.txt'):
+        with open(path) as fp:
+            for line in fp:
+                code, text = line.strip().split(" ", maxsplit=1)
+                ref [code] = text
+    return ref
+
+def get_hypothesis():
+    hyp = {}
+    for path in glob.glob('LibriSpeech/*/*/*/*.flac.txt'):
+        with open(path) as fp:
+            text = fp.read().strip()
+        code = os.path.basename(path).replace('.flac.txt', '')
+        hyp[code] = text
+    return hyp
+
+def get_codes():
+    codes = []
+    for path in glob.glob('LibriSpeech/*/*/*/*.flac'):
+        codes.append(os.path.basename(path).replace('.flac', ''))
+    return sorted(codes)
+
+def main():
+    normalizer = EnglishTextNormalizer()
+
+    ref_orig = get_reference()
+    hyp_orig = get_hypothesis()
+
+    ref_clean = []
+    hyp_clean = []
+
+    for code in get_codes():
+        ref_clean.append(normalizer(ref_orig[code]))
+        hyp_clean.append(normalizer(hyp_orig[code]))
+
+    wer = jiwer.wer(ref_clean, hyp_clean)
+    print(f"WER: {wer * 100:.2f}%")
+
+if __name__ == '__main__':
+    main()
--- a/tests/librispeech/normalizers/LICENSE
+++ b/tests/librispeech/normalizers/LICENSE
@@ -0,0 +1,25 @@
+Code in this directory is adapted from OpenAI Whisper project
+(https://github.com/openai/whisper) and carries the following
+copyright and license.
+
+    MIT License
+
+    Copyright (c) 2022 OpenAI
+
+    Permission is hereby granted, free of charge, to any person obtaining a copy
+    of this software and associated documentation files (the "Software"), to deal
+    in the Software without restriction, including without limitation the rights
+    to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+    copies of the Software, and to permit persons to whom the Software is
+    furnished to do so, subject to the following conditions:
+
+    The above copyright notice and this permission notice shall be included in all
+    copies or substantial portions of the Software.
+
+    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+    OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+    SOFTWARE.
--- a/tests/librispeech/normalizers/init.py
+++ b/tests/librispeech/normalizers/init.py
@@ -0,0 +1,2 @@
+from .basic import BasicTextNormalizer as BasicTextNormalizer
+from .english import EnglishTextNormalizer as EnglishTextNormalizer
--- a/tests/librispeech/normalizers/basic.py
+++ b/tests/librispeech/normalizers/basic.py
@@ -0,0 +1,80 @@
+import re
+import unicodedata
+
+import regex
+
+# non-ASCII letters that are not separated by "NFKD" normalization
+ADDITIONAL_DIACRITICS = {
+    "œ": "oe",
+    "Œ": "OE",
+    "ø": "o",
+    "Ø": "O",
+    "æ": "ae",
+    "Æ": "AE",
+    "ß": "ss",
+    "ẞ": "SS",
+    "đ": "d",
+    "Đ": "D",
+    "ð": "d",
+    "Ð": "D",
+    "þ": "th",
+    "Þ": "th",
+    "ł": "l",
+    "Ł": "L",
+}
+
+
+def remove_symbols_and_diacritics(s: str, keep=""):
+    """
+    Replace any other markers, symbols, and punctuations with a space,
+    and drop any diacritics (category 'Mn' and some manual mappings)
+    """
+    return "".join(
+        (
+            c
+            if c in keep
+            else (
+                ADDITIONAL_DIACRITICS[c]
+                if c in ADDITIONAL_DIACRITICS
+                else (
+                    ""
+                    if unicodedata.category(c) == "Mn"
+                    else " " if unicodedata.category(c)[0] in "MSP" else c
+                )
+            )
+        )
+        for c in unicodedata.normalize("NFKD", s)
+    )
+
+
+def remove_symbols(s: str):
+    """
+    Replace any other markers, symbols, punctuations with a space, keeping diacritics
+    """
+    return "".join(
+        " " if unicodedata.category(c)[0] in "MSP" else c
+        for c in unicodedata.normalize("NFKC", s)
+    )
+
+
+class BasicTextNormalizer:
+    def __init__(self, remove_diacritics: bool = False, split_letters: bool = False):
+        self.clean = (
+            remove_symbols_and_diacritics if remove_diacritics else remove_symbols
+        )
+        self.split_letters = split_letters
+
+    def __call__(self, s: str):
+        s = s.lower()
+        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
+        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
+        s = self.clean(s).lower()
+
+        if self.split_letters:
+            s = " ".join(regex.findall(r"\X", s, regex.U))
+
+        s = re.sub(
+            r"\s+", " ", s
+        )  # replace any successive whitespace characters with a space
+
+        return s
--- a/tests/librispeech/normalizers/english.json
+++ b/tests/librispeech/normalizers/english.json
--- a/tests/librispeech/normalizers/english.py
+++ b/tests/librispeech/normalizers/english.py
@@ -0,0 +1,550 @@
+import json
+import os
+import re
+from fractions import Fraction
+from typing import Iterator, List, Match, Optional, Union
+
+from more_itertools import windowed
+
+from .basic import remove_symbols_and_diacritics
+
+
+class EnglishNumberNormalizer:
+    """
+    Convert any spelled-out numbers into arabic numbers, while handling:
+
+    - remove any commas
+    - keep the suffixes such as: `1960s`, `274th`, `32nd`, etc.
+    - spell out currency symbols after the number. e.g. `$20 million` -> `20000000 dollars`
+    - spell out `one` and `ones`
+    - interpret successive single-digit numbers as nominal: `one oh one` -> `101`
+    """
+
+    def __init__(self):
+        super().__init__()
+
+        self.zeros = {"o", "oh", "zero"}
+        self.ones = {
+            name: i
+            for i, name in enumerate(
+                [
+                    "one",
+                    "two",
+                    "three",
+                    "four",
+                    "five",
+                    "six",
+                    "seven",
+                    "eight",
+                    "nine",
+                    "ten",
+                    "eleven",
+                    "twelve",
+                    "thirteen",
+                    "fourteen",
+                    "fifteen",
+                    "sixteen",
+                    "seventeen",
+                    "eighteen",
+                    "nineteen",
+                ],
+                start=1,
+            )
+        }
+        self.ones_plural = {
+            "sixes" if name == "six" else name + "s": (value, "s")
+            for name, value in self.ones.items()
+        }
+        self.ones_ordinal = {
+            "zeroth": (0, "th"),
+            "first": (1, "st"),
+            "second": (2, "nd"),
+            "third": (3, "rd"),
+            "fifth": (5, "th"),
+            "twelfth": (12, "th"),
+            **{
+                name + ("h" if name.endswith("t") else "th"): (value, "th")
+                for name, value in self.ones.items()
+                if value > 3 and value != 5 and value != 12
+            },
+        }
+        self.ones_suffixed = {**self.ones_plural, **self.ones_ordinal}
+
+        self.tens = {
+            "twenty": 20,
+            "thirty": 30,
+            "forty": 40,
+            "fifty": 50,
+            "sixty": 60,
+            "seventy": 70,
+            "eighty": 80,
+            "ninety": 90,
+        }
+        self.tens_plural = {
+            name.replace("y", "ies"): (value, "s") for name, value in self.tens.items()
+        }
+        self.tens_ordinal = {
+            name.replace("y", "ieth"): (value, "th")
+            for name, value in self.tens.items()
+        }
+        self.tens_suffixed = {**self.tens_plural, **self.tens_ordinal}
+
+        self.multipliers = {
+            "hundred": 100,
+            "thousand": 1_000,
+            "million": 1_000_000,
+            "billion": 1_000_000_000,
+            "trillion": 1_000_000_000_000,
+            "quadrillion": 1_000_000_000_000_000,
+            "quintillion": 1_000_000_000_000_000_000,
+            "sextillion": 1_000_000_000_000_000_000_000,
+            "septillion": 1_000_000_000_000_000_000_000_000,
+            "octillion": 1_000_000_000_000_000_000_000_000_000,
+            "nonillion": 1_000_000_000_000_000_000_000_000_000_000,
+            "decillion": 1_000_000_000_000_000_000_000_000_000_000_000,
+        }
+        self.multipliers_plural = {
+            name + "s": (value, "s") for name, value in self.multipliers.items()
+        }
+        self.multipliers_ordinal = {
+            name + "th": (value, "th") for name, value in self.multipliers.items()
+        }
+        self.multipliers_suffixed = {
+            **self.multipliers_plural,
+            **self.multipliers_ordinal,
+        }
+        self.decimals = {*self.ones, *self.tens, *self.zeros}
+
+        self.preceding_prefixers = {
+            "minus": "-",
+            "negative": "-",
+            "plus": "+",
+            "positive": "+",
+        }
+        self.following_prefixers = {
+            "pound": "£",
+            "pounds": "£",
+            "euro": "€",
+            "euros": "€",
+            "dollar": "$",
+            "dollars": "$",
+            "cent": "¢",
+            "cents": "¢",
+        }
+        self.prefixes = set(
+            list(self.preceding_prefixers.values())
+            + list(self.following_prefixers.values())
+        )
+        self.suffixers = {
+            "per": {"cent": "%"},
+            "percent": "%",
+        }
+        self.specials = {"and", "double", "triple", "point"}
+
+        self.words = set(
+            [
+                key
+                for mapping in [
+                    self.zeros,
+                    self.ones,
+                    self.ones_suffixed,
+                    self.tens,
+                    self.tens_suffixed,
+                    self.multipliers,
+                    self.multipliers_suffixed,
+                    self.preceding_prefixers,
+                    self.following_prefixers,
+                    self.suffixers,
+                    self.specials,
+                ]
+                for key in mapping
+            ]
+        )
+        self.literal_words = {"one", "ones"}
+
+    def process_words(self, words: List[str]) -> Iterator[str]:
+        prefix: Optional[str] = None
+        value: Optional[Union[str, int]] = None
+        skip = False
+
+        def to_fraction(s: str):
+            try:
+                return Fraction(s)
+            except ValueError:
+                return None
+
+        def output(result: Union[str, int]):
+            nonlocal prefix, value
+            result = str(result)
+            if prefix is not None:
+                result = prefix + result
+            value = None
+            prefix = None
+            return result
+
+        if len(words) == 0:
+            return
+
+        for prev, current, next in windowed([None] + words + [None], 3):
+            if skip:
+                skip = False
+                continue
+
+            next_is_numeric = next is not None and re.match(r"^\d+(\.\d+)?$", next)
+            has_prefix = current[0] in self.prefixes
+            current_without_prefix = current[1:] if has_prefix else current
+            if re.match(r"^\d+(\.\d+)?$", current_without_prefix):
+                # arabic numbers (potentially with signs and fractions)
+                f = to_fraction(current_without_prefix)
+                assert f is not None
+                if value is not None:
+                    if isinstance(value, str) and value.endswith("."):
+                        # concatenate decimals / ip address components
+                        value = str(value) + str(current)
+                        continue
+                    else:
+                        yield output(value)
+
+                prefix = current[0] if has_prefix else prefix
+                if f.denominator == 1:
+                    value = f.numerator  # store integers as int
+                else:
+                    value = current_without_prefix
+            elif current not in self.words:
+                # non-numeric words
+                if value is not None:
+                    yield output(value)
+                yield output(current)
+            elif current in self.zeros:
+                value = str(value or "") + "0"
+            elif current in self.ones:
+                ones = self.ones[current]
+
+                if value is None:
+                    value = ones
+                elif isinstance(value, str) or prev in self.ones:
+                    if (
+                        prev in self.tens and ones < 10
+                    ):  # replace the last zero with the digit
+                        assert value[-1] == "0"
+                        value = value[:-1] + str(ones)
+                    else:
+                        value = str(value) + str(ones)
+                elif ones < 10:
+                    if value % 10 == 0:
+                        value += ones
+                    else:
+                        value = str(value) + str(ones)
+                else:  # eleven to nineteen
+                    if value % 100 == 0:
+                        value += ones
+                    else:
+                        value = str(value) + str(ones)
+            elif current in self.ones_suffixed:
+                # ordinal or cardinal; yield the number right away
+                ones, suffix = self.ones_suffixed[current]
+                if value is None:
+                    yield output(str(ones) + suffix)
+                elif isinstance(value, str) or prev in self.ones:
+                    if prev in self.tens and ones < 10:
+                        assert value[-1] == "0"
+                        yield output(value[:-1] + str(ones) + suffix)
+                    else:
+                        yield output(str(value) + str(ones) + suffix)
+                elif ones < 10:
+                    if value % 10 == 0:
+                        yield output(str(value + ones) + suffix)
+                    else:
+                        yield output(str(value) + str(ones) + suffix)
+                else:  # eleven to nineteen
+                    if value % 100 == 0:
+                        yield output(str(value + ones) + suffix)
+                    else:
+                        yield output(str(value) + str(ones) + suffix)
+                value = None
+            elif current in self.tens:
+                tens = self.tens[current]
+                if value is None:
+                    value = tens
+                elif isinstance(value, str):
+                    value = str(value) + str(tens)
+                else:
+                    if value % 100 == 0:
+                        value += tens
+                    else:
+                        value = str(value) + str(tens)
+            elif current in self.tens_suffixed:
+                # ordinal or cardinal; yield the number right away
+                tens, suffix = self.tens_suffixed[current]
+                if value is None:
+                    yield output(str(tens) + suffix)
+                elif isinstance(value, str):
+                    yield output(str(value) + str(tens) + suffix)
+                else:
+                    if value % 100 == 0:
+                        yield output(str(value + tens) + suffix)
+                    else:
+                        yield output(str(value) + str(tens) + suffix)
+            elif current in self.multipliers:
+                multiplier = self.multipliers[current]
+                if value is None:
+                    value = multiplier
+                elif isinstance(value, str) or value == 0:
+                    f = to_fraction(value)
+                    p = f * multiplier if f is not None else None
+                    if f is not None and p.denominator == 1:
+                        value = p.numerator
+                    else:
+                        yield output(value)
+                        value = multiplier
+                else:
+                    before = value // 1000 * 1000
+                    residual = value % 1000
+                    value = before + residual * multiplier
+            elif current in self.multipliers_suffixed:
+                multiplier, suffix = self.multipliers_suffixed[current]
+                if value is None:
+                    yield output(str(multiplier) + suffix)
+                elif isinstance(value, str):
+                    f = to_fraction(value)
+                    p = f * multiplier if f is not None else None
+                    if f is not None and p.denominator == 1:
+                        yield output(str(p.numerator) + suffix)
+                    else:
+                        yield output(value)
+                        yield output(str(multiplier) + suffix)
+                else:  # int
+                    before = value // 1000 * 1000
+                    residual = value % 1000
+                    value = before + residual * multiplier
+                    yield output(str(value) + suffix)
+                value = None
+            elif current in self.preceding_prefixers:
+                # apply prefix (positive, minus, etc.) if it precedes a number
+                if value is not None:
+                    yield output(value)
+
+                if next in self.words or next_is_numeric:
+                    prefix = self.preceding_prefixers[current]
+                else:
+                    yield output(current)
+            elif current in self.following_prefixers:
+                # apply prefix (dollars, cents, etc.) only after a number
+                if value is not None:
+                    prefix = self.following_prefixers[current]
+                    yield output(value)
+                else:
+                    yield output(current)
+            elif current in self.suffixers:
+                # apply suffix symbols (percent -> '%')
+                if value is not None:
+                    suffix = self.suffixers[current]
+                    if isinstance(suffix, dict):
+                        if next in suffix:
+                            yield output(str(value) + suffix[next])
+                            skip = True
+                        else:
+                            yield output(value)
+                            yield output(current)
+                    else:
+                        yield output(str(value) + suffix)
+                else:
+                    yield output(current)
+            elif current in self.specials:
+                if next not in self.words and not next_is_numeric:
+                    # apply special handling only if the next word can be numeric
+                    if value is not None:
+                        yield output(value)
+                    yield output(current)
+                elif current == "and":
+                    # ignore "and" after hundreds, thousands, etc.
+                    if prev not in self.multipliers:
+                        if value is not None:
+                            yield output(value)
+                        yield output(current)
+                elif current == "double" or current == "triple":
+                    if next in self.ones or next in self.zeros:
+                        repeats = 2 if current == "double" else 3
+                        ones = self.ones.get(next, 0)
+                        value = str(value or "") + str(ones) * repeats
+                        skip = True
+                    else:
+                        if value is not None:
+                            yield output(value)
+                        yield output(current)
+                elif current == "point":
+                    if next in self.decimals or next_is_numeric:
+                        value = str(value or "") + "."
+                else:
+                    # should all have been covered at this point
+                    raise ValueError(f"Unexpected token: {current}")
+            else:
+                # all should have been covered at this point
+                raise ValueError(f"Unexpected token: {current}")
+
+        if value is not None:
+            yield output(value)
+
+    def preprocess(self, s: str):
+        # replace "<number> and a half" with "<number> point five"
+        results = []
+
+        segments = re.split(r"\band\s+a\s+half\b", s)
+        for i, segment in enumerate(segments):
+            if len(segment.strip()) == 0:
+                continue
+            if i == len(segments) - 1:
+                results.append(segment)
+            else:
+                results.append(segment)
+                last_word = segment.rsplit(maxsplit=2)[-1]
+                if last_word in self.decimals or last_word in self.multipliers:
+                    results.append("point five")
+                else:
+                    results.append("and a half")
+
+        s = " ".join(results)
+
+        # put a space at number/letter boundary
+        s = re.sub(r"([a-z])([0-9])", r"\1 \2", s)
+        s = re.sub(r"([0-9])([a-z])", r"\1 \2", s)
+
+        # but remove spaces which could be a suffix
+        s = re.sub(r"([0-9])\s+(st|nd|rd|th|s)\b", r"\1\2", s)
+
+        return s
+
+    def postprocess(self, s: str):
+        def combine_cents(m: Match):
+            try:
+                currency = m.group(1)
+                integer = m.group(2)
+                cents = int(m.group(3))
+                return f"{currency}{integer}.{cents:02d}"
+            except ValueError:
+                return m.string
+
+        def extract_cents(m: Match):
+            try:
+                return f"¢{int(m.group(1))}"
+            except ValueError:
+                return m.string
+
+        # apply currency postprocessing; "$2 and ¢7" -> "$2.07"
+        s = re.sub(r"([€£$])([0-9]+) (?:and )?¢([0-9]{1,2})\b", combine_cents, s)
+        s = re.sub(r"[€£$]0.([0-9]{1,2})\b", extract_cents, s)
+
+        # write "one(s)" instead of "1(s)", just for the readability
+        s = re.sub(r"\b1(s?)\b", r"one\1", s)
+
+        return s
+
+    def __call__(self, s: str):
+        s = self.preprocess(s)
+        s = " ".join(word for word in self.process_words(s.split()) if word is not None)
+        s = self.postprocess(s)
+
+        return s
+
+
+class EnglishSpellingNormalizer:
+    """
+    Applies British-American spelling mappings as listed in [1].
+
+    [1] https://www.tysto.com/uk-us-spelling-list.html
+    """
+
+    def __init__(self):
+        mapping_path = os.path.join(os.path.dirname(__file__), "english.json")
+        self.mapping = json.load(open(mapping_path))
+
+    def __call__(self, s: str):
+        return " ".join(self.mapping.get(word, word) for word in s.split())
+
+
+class EnglishTextNormalizer:
+    def __init__(self):
+        self.ignore_patterns = r"\b(hmm|mm|mhm|mmm|uh|um)\b"
+        self.replacers = {
+            # common contractions
+            r"\bwon't\b": "will not",
+            r"\bcan't\b": "can not",
+            r"\blet's\b": "let us",
+            r"\bain't\b": "aint",
+            r"\by'all\b": "you all",
+            r"\bwanna\b": "want to",
+            r"\bgotta\b": "got to",
+            r"\bgonna\b": "going to",
+            r"\bi'ma\b": "i am going to",
+            r"\bimma\b": "i am going to",
+            r"\bwoulda\b": "would have",
+            r"\bcoulda\b": "could have",
+            r"\bshoulda\b": "should have",
+            r"\bma'am\b": "madam",
+            # contractions in titles/prefixes
+            r"\bmr\b": "mister ",
+            r"\bmrs\b": "missus ",
+            r"\bst\b": "saint ",
+            r"\bdr\b": "doctor ",
+            r"\bprof\b": "professor ",
+            r"\bcapt\b": "captain ",
+            r"\bgov\b": "governor ",
+            r"\bald\b": "alderman ",
+            r"\bgen\b": "general ",
+            r"\bsen\b": "senator ",
+            r"\brep\b": "representative ",
+            r"\bpres\b": "president ",
+            r"\brev\b": "reverend ",
+            r"\bhon\b": "honorable ",
+            r"\basst\b": "assistant ",
+            r"\bassoc\b": "associate ",
+            r"\blt\b": "lieutenant ",
+            r"\bcol\b": "colonel ",
+            r"\bjr\b": "junior ",
+            r"\bsr\b": "senior ",
+            r"\besq\b": "esquire ",
+            # prefect tenses, ideally it should be any past participles, but it's harder..
+            r"'d been\b": " had been",
+            r"'s been\b": " has been",
+            r"'d gone\b": " had gone",
+            r"'s gone\b": " has gone",
+            r"'d done\b": " had done",  # "'s done" is ambiguous
+            r"'s got\b": " has got",
+            # general contractions
+            r"n't\b": " not",
+            r"'re\b": " are",
+            r"'s\b": " is",
+            r"'d\b": " would",
+            r"'ll\b": " will",
+            r"'t\b": " not",
+            r"'ve\b": " have",
+            r"'m\b": " am",
+        }
+        self.standardize_numbers = EnglishNumberNormalizer()
+        self.standardize_spellings = EnglishSpellingNormalizer()
+
+    def __call__(self, s: str):
+        s = s.lower()
+
+        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
+        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
+        s = re.sub(self.ignore_patterns, "", s)
+        s = re.sub(r"\s+'", "'", s)  # when there's a space before an apostrophe
+
+        for pattern, replacement in self.replacers.items():
+            s = re.sub(pattern, replacement, s)
+
+        s = re.sub(r"(\d),(\d)", r"\1\2", s)  # remove commas between digits
+        s = re.sub(r"\.([^0-9]|$)", r" \1", s)  # remove periods not followed by numbers
+        s = remove_symbols_and_diacritics(s, keep=".%$¢€£")  # keep numeric symbols
+
+        s = self.standardize_numbers(s)
+        s = self.standardize_spellings(s)
+
+        # now remove prefix/suffix symbols that are not preceded/followed by numbers
+        s = re.sub(r"[.$¢€£]([^0-9])", r" \1", s)
+        s = re.sub(r"([^0-9])%", r"\1 ", s)
+
+        s = re.sub(r"\s+", " ", s)  # replace any successive whitespaces with a space
+
+        return s
--- a/tests/parakeet-expected-diffusion-output.txt
+++ b/tests/parakeet-expected-diffusion-output.txt
--- a/tests/parakeet-expected-gb1-output.txt
+++ b/tests/parakeet-expected-gb1-output.txt
@@ -0,0 +1 @@
+My fellow Americans, this day has brought terrible news and great sadness to our country. At nine o'clock this morning, mission control in Houston lost contact with our space shuttle Columbia. A short time later, debris was seen falling from the skies above Texas. The Columbia's lost. There are no survivors. On board was a crew of seven. Colonel Rick Husband, Lieutenant Colonel Michael Anderson, Commander Laurel Clark, Captain David Brown, Commander William McCool, Dr. Kulpna Shavla, and Ilan Ramon, a colonel in the Israeli Air Force. These men and women assumed great risk in the service to all humanity. In an age when space flight has come to seem almost routine. It is easy to overlook the dangers of travel by rocket and the difficulties of navigating the fierce outer atmosphere of the earth. These astronauts knew the dangers, and they faced them willingly, knowing they had a high and noble purpose in life. Because of their courage and daring and idealism, we will miss them all the more. And those you loved will always have the respect and gratitude of this country. The cause in which they died will continue. Mankind is led into the darkness beyond our world by the inspiration of discovery and the longing to understand. Our journey into space will go on. In the skies today, we saw destruction and tragedy. Yet farther than we can see, there is comfort and hope. In the words of the prophet Isaiah, lift your eyes and look to the heavens. Who created all these? He who brings out the starry hosts one by one and calls them each by name. Because of his great power and mighty strength, not one of them is missing. The same creator who names the stars also knows the names of the seven souls we mourn today. The crew of the shuttle Columbia did not return safely to Earth. Yet we can pray that all are safely home. May God bless the grieving families and make out may God continue to bless America.
--- a/tests/parakeet-expected-jfk-output.txt
+++ b/tests/parakeet-expected-jfk-output.txt
@@ -0,0 +1 @@
+And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.
--- a/tests/parakeet-verification.h
+++ b/tests/parakeet-verification.h
@@ -0,0 +1,110 @@
+#pragma once
+
+#include <algorithm>
+#include <cassert>
+#include <cctype>
+#include <cstdio>
+#include <fstream>
+#include <iterator>
+#include <string>
+#include <vector>
+
+#ifndef TRANSCRIPTION_SIMILARITY_THRESHOLD
+#define TRANSCRIPTION_SIMILARITY_THRESHOLD 1.0
+#endif
+
+static std::string read_expected_transcription(const char * path) {
+    std::ifstream fin(path);
+    assert(fin.is_open());
+
+    std::string text(
+        (std::istreambuf_iterator<char>(fin)),
+         std::istreambuf_iterator<char>());
+
+    while (!text.empty() && (text.back() == '\n' || text.back() == '\r')) {
+        text.pop_back();
+    }
+
+    return text;
+}
+
+static std::vector<std::string> transcription_words(const std::string & text) {
+    std::vector<std::string> words;
+    std::string word;
+
+    for (unsigned char ch : text) {
+        if (std::isalnum(ch)) {
+            word.push_back((char) std::tolower(ch));
+        } else if (!word.empty()) {
+            words.push_back(word);
+            word.clear();
+        }
+    }
+
+    if (!word.empty()) {
+        words.push_back(word);
+    }
+
+    return words;
+}
+
+static double transcription_lcs_similarity(const std::string & expected, const std::string & actual) {
+    const std::vector<std::string> expected_words = transcription_words(expected);
+    const std::vector<std::string> actual_words   = transcription_words(actual);
+
+    if (expected_words.empty() && actual_words.empty()) {
+        return 1.0;
+    }
+
+    if (expected_words.empty() || actual_words.empty()) {
+        return 0.0;
+    }
+
+    std::vector<int> prev(actual_words.size() + 1, 0);
+    std::vector<int> cur (actual_words.size() + 1, 0);
+
+    for (size_t i = 0; i < expected_words.size(); ++i) {
+        std::fill(cur.begin(), cur.end(), 0);
+
+        for (size_t j = 0; j < actual_words.size(); ++j) {
+            if (expected_words[i] == actual_words[j]) {
+                cur[j + 1] = prev[j] + 1;
+            } else {
+                cur[j + 1] = std::max(prev[j + 1], cur[j]);
+            }
+        }
+
+        prev.swap(cur);
+    }
+
+    const int lcs = prev[actual_words.size()];
+    return (2.0 * lcs) / (expected_words.size() + actual_words.size());
+}
+
+static bool verify_transcription(const std::string & expected, const std::string & actual) {
+    const double threshold = TRANSCRIPTION_SIMILARITY_THRESHOLD;
+
+    if (threshold >= 1.0) {
+        if (actual == expected) {
+            return true;
+        }
+
+        fprintf(stderr, "\n\n");
+        fprintf(stderr, "[Failed] Transcript mismatched\n");
+        fprintf(stderr, "expected:\n%s\n\n", expected.c_str());
+        fprintf(stderr, "actual:\n%s\n", actual.c_str());
+        return false;
+    }
+
+    const double similarity = transcription_lcs_similarity(expected, actual);
+    printf("\nTranscript similarity: %.6f (threshold %.6f)\n", similarity, threshold);
+
+    if (similarity >= threshold) {
+        return true;
+    }
+
+    fprintf(stderr, "\n\nTranscript similarity below threshold: %.6f < %.6f\n", similarity, threshold);
+    fprintf(stderr, "Expected:\n%s\n\n", expected.c_str());
+    fprintf(stderr, "Actual:\n%s\n", actual.c_str());
+    return false;
+}
--- a/tests/run-tests.sh
+++ b/tests/run-tests.sh
@@ -0,0 +1,164 @@
+#!/bin/bash
+
+# This scripts run the selected model agains a collection of audio files from the web.
+# It downloads, converts and transcribes each file and then compares the result with the expected reference
+# transcription. The comparison is performed using git's diff command and shows the differences at the character level.
+# It can be used to quickly verify that the model is working as expected across a wide range of audio files.
+# I.e. like an integration test. The verification is done by visual inspection of the diff output.
+#
+# The reference data can be for example generated using the original OpenAI Whisper implementation, or entered manually.
+#
+# Feel free to suggest extra audio files to add to the list.
+# Make sure they are between 1-3 minutes long since we don't want to make the test too slow.
+#
+# Usage:
+#
+#   ./tests/run-tests.sh <model_name> [threads]
+#
+
+cd `dirname $0`
+
+# Whisper models
+models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large-v2" "large-v3" "large-v3-turbo" )
+
+# Parakeet model variants
+parakeet_models=( "f16" "f32" "q2_k" "q4_0" "q4_k" "q8_0" )
+
+# list available models
+function list_models {
+    printf "\n"
+    printf "  Available whisper models:"
+    for model in "${models[@]}"; do
+        printf " $model"
+    done
+    printf "\n"
+    printf "  Available parakeet models:"
+    for model in "${parakeet_models[@]}"; do
+        printf " parakeet-$model"
+    done
+    printf "\n\n"
+}
+
+if [ $# -eq 0 ]; then
+    printf "Usage: $0 [model] [threads]\n\n"
+    printf "No model specified. Aborting\n"
+    list_models
+    exit 1
+fi
+
+model=$1
+
+threads=""
+if [ $# -eq 2 ]; then
+    threads="-t $2"
+fi
+
+# Detect parakeet model (prefix "parakeet-" or a bare variant like "f32")
+is_parakeet=0
+parakeet_variant=""
+if [[ $model == parakeet-* ]]; then
+    is_parakeet=1
+    parakeet_variant="${model#parakeet-}"
+fi
+for v in "${parakeet_models[@]}"; do
+    if [[ $model == "$v" ]]; then
+        is_parakeet=1
+        parakeet_variant="$v"
+        break
+    fi
+done
+
+if [ $is_parakeet -eq 1 ]; then
+    main="../build/bin/parakeet-cli"
+    model_path="../models/ggml-parakeet-tdt-0.6b-v3-${parakeet_variant}.bin"
+else
+    main="../build/bin/whisper-cli"
+    model_path="../models/ggml-${model}.bin"
+fi
+
+if [ ! -f $model_path ]; then
+    printf "Model $model not found ($model_path). Aborting\n"
+    list_models
+    exit 1
+fi
+
+if [ ! -f $main ]; then
+    printf "Executable $main not found. Aborting\n"
+    exit 1
+fi
+
+# add various audio files for testing purposes here
+# the order of the files is important so don't change the existing order
+# when adding new files, make sure to add the expected "ref.txt" file with the correct transcript
+urls_en=(
+    "https://upload.wikimedia.org/wikipedia/commons/1/1f/George_W_Bush_Columbia_FINAL.ogg"
+    "https://upload.wikimedia.org/wikipedia/en/d/d4/En.henryfphillips.ogg"
+    "https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav"
+)
+
+urls_es=(
+    "https://upload.wikimedia.org/wikipedia/commons/c/c1/La_contaminacion_del_agua.ogg"
+)
+
+urls_it=(
+)
+
+urls_pt=(
+)
+
+urls_de=(
+)
+
+urls_jp=(
+)
+
+urls_ru=(
+)
+
+function run_lang() {
+    lang=$1
+    shift
+    urls=("$@")
+
+    i=0
+    for url in "${urls[@]}"; do
+        echo "- [$lang] Processing '$url' ..."
+
+        ext="${url##*.}"
+        fname_src="$lang-${i}.${ext}"
+        fname_dst="$lang-${i}-16khz.wav"
+
+        if [ ! -f $fname_src ]; then
+            wget --quiet --show-progress -O $fname_src $url
+        fi
+
+        if [ ! -f $fname_dst ]; then
+            ffmpeg -loglevel -0 -y -i $fname_src -ar 16000 -ac 1 -c:a pcm_s16le $fname_dst
+            if [ $? -ne 0 ]; then
+                echo "Error: ffmpeg failed to convert $fname_src to $fname_dst"
+                exit 1
+            fi
+        fi
+
+        if [ $is_parakeet -eq 1 ]; then
+            $main -m $model_path $threads -f $fname_dst -otxt 2> /dev/null
+        else
+            $main -m $model_path $threads -f $fname_dst -l $lang -otxt 2> /dev/null
+        fi
+
+        git diff --no-index --word-diff=color --word-diff-regex=. $lang-$i-ref.txt $fname_dst.txt
+
+        i=$(($i+1))
+    done
+}
+
+run_lang "en" "${urls_en[@]}"
+
+if [ $is_parakeet -eq 0 ] && [[ $model != *.en* ]]; then
+    run_lang "es" "${urls_es[@]}"
+    run_lang "it" "${urls_it[@]}"
+    run_lang "pt" "${urls_pt[@]}"
+    run_lang "de" "${urls_de[@]}"
+    run_lang "jp" "${urls_jp[@]}"
+    run_lang "ru" "${urls_ru[@]}"
+fi
--- a/tests/test-c.c
+++ b/tests/test-c.c
@@ -0,0 +1,3 @@
+#include "whisper.h"
+
+int main(void) {}
--- a/tests/test-common-utf8.cpp
+++ b/tests/test-common-utf8.cpp
@@ -0,0 +1,34 @@
+#include "common-whisper.h"
+
+#include <cstdlib>
+#include <cstdio>
+#include <string>
+
+static void expect_needed(const std::string & input, int expected) {
+    const int actual = utf8_trailing_bytes_needed(input);
+    if (actual != expected) {
+        fprintf(stderr, "expected %d trailing UTF-8 bytes, got %d\n", expected, actual);
+        std::abort();
+    }
+}
+
+int main() {
+    expect_needed("", 0);
+    expect_needed("plain ascii", 0);
+
+    const std::string cjk = "\xE4\xBD\xA0"; // U+4F60
+    expect_needed(cjk.substr(0, 1), 2);
+    expect_needed(cjk.substr(0, 2), 1);
+    expect_needed(cjk, 0);
+
+    const std::string emoji = "\xF0\x9F\x98\x80"; // U+1F600
+    expect_needed(emoji.substr(0, 1), 3);
+    expect_needed(emoji.substr(0, 2), 2);
+    expect_needed(emoji.substr(0, 3), 1);
+    expect_needed(emoji, 0);
+
+    expect_needed("\x80\x80", 0);
+    expect_needed("\xFF", 0);
+
+    return 0;
+}
--- a/tests/test-parakeet-full.cpp
+++ b/tests/test-parakeet-full.cpp
@@ -0,0 +1,101 @@
+#include "parakeet.h"
+#include "common-whisper.h"
+#include "parakeet-verification.h"
+
+#include <cstdio>
+#include <string>
+
+#ifdef NDEBUG
+#undef NDEBUG
+#endif
+#include <cassert>
+
+struct test_state {
+    bool is_first = true;
+    std::string transcript;
+};
+
+void progress_callback(parakeet_context * ctx, parakeet_state * state, int progress, void * user_data) {
+    bool * called = static_cast<bool *>(user_data);
+    *called = true;
+}
+
+bool encoder_begin_callback(parakeet_context * ctx, parakeet_state * state, void * user_data) {
+    bool * called = static_cast<bool *>(user_data);
+    *called = true;
+    return true;
+}
+
+bool abort_callback(void * user_data) {
+    bool * called = static_cast<bool *>(user_data);
+    *called = true;
+    return false; // just continue without aborting.
+}
+
+void token_callback(parakeet_context * ctx, parakeet_state * state, const parakeet_token_data * token_data, void * user_data) {
+    test_state * tstate = static_cast<test_state *>(user_data);
+
+    const char * token_str = parakeet_token_to_str(ctx, token_data->id);
+    char text_buf[256];
+    parakeet_token_to_text(token_str, tstate->is_first, text_buf, sizeof(text_buf));
+
+    printf("%s", text_buf);
+    fflush(stdout);
+
+    tstate->transcript += text_buf;
+    tstate->is_first = false;
+}
+
+int main() {
+    std::string model_path  = PARAKEET_MODEL_PATH;
+    std::string sample_path = SAMPLE_PATH;
+
+    std::vector<float> pcmf32;
+    std::vector<std::vector<float>> pcmf32s;
+    assert(read_audio_data(sample_path.c_str(), pcmf32, pcmf32s, false));
+    assert(pcmf32.size() > 0);
+    assert(pcmf32s.size() == 0); // no stereo vector
+
+    printf("Loading Parakeet model from: %s\n", model_path.c_str());
+
+    struct parakeet_context_params ctx_params = parakeet_context_default_params();
+
+    struct parakeet_context * pctx = parakeet_init_from_file_with_params(model_path.c_str(), ctx_params);
+    if (pctx == nullptr) {
+        fprintf(stderr, "Failed to load Parakeet model\n");
+        return 1;
+    }
+    printf("Successfully loaded Parakeet model\n");
+
+    struct parakeet_full_params params = parakeet_full_default_params(PARAKEET_SAMPLING_GREEDY);
+    test_state tstate;
+    params.new_token_callback = token_callback;
+    params.new_token_callback_user_data = &tstate;
+    bool progress_callback_called = false;
+    params.progress_callback = progress_callback;
+    params.progress_callback_user_data = &progress_callback_called;
+    bool encoder_begin_callback_called = false;
+    params.encoder_begin_callback = encoder_begin_callback;
+    params.encoder_begin_callback_user_data = &encoder_begin_callback_called;
+    bool abort_callback_called = false;
+    params.abort_callback = abort_callback;
+    params.abort_callback_user_data = &abort_callback_called;
+
+    int ret = parakeet_full(pctx, params, pcmf32.data(), pcmf32.size());
+    assert(ret == 0);
+    assert(progress_callback_called);
+    assert(encoder_begin_callback_called);
+    assert(abort_callback_called);
+
+    const std::string expected = read_expected_transcription(EXPECTED_TRANSCRIPTION_PATH);
+    const bool transcript_matches = verify_transcription(expected, tstate.transcript);
+
+    parakeet_free(pctx);
+
+    if (!transcript_matches) {
+        return 1;
+    }
+
+    printf("\nTest passed: parakeet_full succeeded!\n");
+    return 0;
+}
--- a/tests/test-parakeet.cpp
+++ b/tests/test-parakeet.cpp
@@ -0,0 +1,126 @@
+#include "parakeet.h"
+#include "common-whisper.h"
+
+#include <cstdio>
+#include <string>
+
+#ifdef NDEBUG
+#undef NDEBUG
+#endif
+#include <cassert>
+
+void token_callback(parakeet_context * ctx, parakeet_state * state, const parakeet_token_data * token_data, void * user_data) {
+    static bool is_first = true;
+    const char * token_str = parakeet_token_to_str(ctx, token_data->id);
+    char text_buf[256];
+    parakeet_token_to_text(token_str, is_first, text_buf, sizeof(text_buf));
+
+    int32_t time_ms = token_data->frame_index * 10;
+
+    printf("%s", text_buf);
+    fflush(stdout);
+
+    is_first = false;
+}
+
+void segment_callback(parakeet_context * ctx, parakeet_state * state, int n_new, void * user_data) {
+    const int n_segments = parakeet_full_n_segments_from_state(state);
+    const int s0 = n_segments - n_new;
+
+    printf("\nSegment Callback: %d new segment(s)\n", n_new);
+
+    for (int i = s0; i < n_segments; i++) {
+        const char * text = parakeet_full_get_segment_text_from_state(state, i);
+        const int64_t t0 = parakeet_full_get_segment_t0_from_state(state, i);
+        const int64_t t1 = parakeet_full_get_segment_t1_from_state(state, i);
+
+        printf("Segment %d: [%lld -> %lld] \"%s\"\n", i, (long long)t0, (long long)t1, text);
+        printf("Tokens:\n");
+
+        const int n_tokens = parakeet_full_n_tokens_from_state(state, i);
+        for (int j = 0; j < n_tokens; j++) {
+            parakeet_token_data token_data = parakeet_full_get_token_data_from_state(state, i, j);
+            const char * token_str = parakeet_token_to_str(ctx, token_data.id);
+
+            printf("  [%2d] id=%5d frame=%3d dur_idx=%2d dur_val=%2d p=%.4f plog=%.4f t0=%4lld t1=%4lld word_start=%d \"%s\"\n",
+                   j,
+                   token_data.id,
+                   token_data.frame_index,
+                   token_data.duration_idx,
+                   token_data.duration_value,
+                   token_data.p,
+                   token_data.plog,
+                   (long long)token_data.t0,
+                   (long long)token_data.t1,
+                   token_data.is_word_start,
+                   token_str);
+        }
+    }
+    printf("\n");
+}
+
+static int test_invalid_model_load(){
+    struct parakeet_context_params ctx_params = parakeet_context_default_params();
+    struct parakeet_context * pctx =
+        parakeet_init_from_file_with_params_no_state(PARAKEET_BAD_MODEL_PATH, ctx_params);
+    if(pctx != nullptr){
+        fprintf(stderr, "Expected invalid Parakeet model to fail loading \n");
+        parakeet_free(pctx);
+        return 1;
+    }
+    return 0;
+}
+
+static int test_valid_model() {
+    std::string model_path  = PARAKEET_MODEL_PATH;
+    std::string sample_path = SAMPLE_PATH;
+
+    // Load the sample audio file
+    std::vector<float> pcmf32;
+    std::vector<std::vector<float>> pcmf32s;
+    assert(read_audio_data(sample_path.c_str(), pcmf32, pcmf32s, false));
+    assert(pcmf32.size() > 0);
+    assert(pcmf32s.size() == 0);
+
+    printf("Loading Parakeet model from: %s\n", model_path.c_str());
+
+    struct parakeet_context_params ctx_params = parakeet_context_default_params();
+
+    struct parakeet_context * pctx = parakeet_init_from_file_with_params_no_state(model_path.c_str(), ctx_params);
+    if (pctx == nullptr) {
+        fprintf(stderr, "Failed to load Parakeet model\n");
+        return 1;
+    }
+    printf("Successfully loaded Parakeet model\n");
+
+    struct parakeet_full_params params = parakeet_full_default_params(PARAKEET_SAMPLING_GREEDY);
+    params.new_token_callback = token_callback;
+    params.new_token_callback_user_data = nullptr;
+    params.new_segment_callback = segment_callback;
+    params.new_segment_callback_user_data = nullptr;
+    parakeet_state * state = parakeet_init_state(pctx);
+
+    int ret = parakeet_chunk(pctx, state, params, pcmf32.data(), pcmf32.size());
+    assert(ret == 0);
+
+    parakeet_free_state(state);
+    parakeet_free(pctx);
+
+    printf("\nTest passed: Parakeet model loaded and freed successfully\n");
+    return 0;
+}
+
+int main(){
+    ggml_backend_load_all();
+
+    if(test_valid_model() != 0){
+        return 1;
+    }
+
+    if(test_invalid_model_load() != 0){
+        return 1;
+    }
+
+    printf("\nTest passed: Parakeet model load tests completed successfully\n");
+    return 0;
+}
--- a/tests/test-vad-full.cpp
+++ b/tests/test-vad-full.cpp
@@ -0,0 +1,87 @@
+#include "whisper.h"
+#include "common-whisper.h"
+
+#include <cstdio>
+#include <cfloat>
+#include <string>
+#include <cstring>
+
+#ifdef NDEBUG
+#undef NDEBUG
+#endif
+
+#include <cassert>
+
+int main() {
+    ggml_backend_load_all();
+
+    std::string whisper_model_path = WHISPER_MODEL_PATH;
+    std::string vad_model_path     = VAD_MODEL_PATH;
+    std::string sample_path        = SAMPLE_PATH;
+
+    // Load the sample audio file
+    std::vector<float> pcmf32;
+    std::vector<std::vector<float>> pcmf32s;
+    assert(read_audio_data(sample_path.c_str(), pcmf32, pcmf32s, false));
+
+    struct whisper_context_params cparams = whisper_context_default_params();
+    struct whisper_context * wctx = whisper_init_from_file_with_params(
+            whisper_model_path.c_str(),
+            cparams);
+
+    if (!wctx) {
+        fprintf(stderr, "failed to load model '%s'\n", whisper_model_path.c_str());
+        return 1;
+    }
+
+    struct whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_BEAM_SEARCH);
+    wparams.vad            = true;
+    wparams.vad_model_path = vad_model_path.c_str();
+
+    wparams.vad_params.threshold               = 0.5f;
+    wparams.vad_params.min_speech_duration_ms  = 250;
+    wparams.vad_params.min_silence_duration_ms = 100;
+    wparams.vad_params.max_speech_duration_s   = FLT_MAX;
+    wparams.vad_params.speech_pad_ms           = 30;
+
+    assert(whisper_full_parallel(wctx, wparams, pcmf32.data(), pcmf32.size(), 1) == 0);
+
+    const int n_segments = whisper_full_n_segments(wctx);
+    assert(n_segments == 1);
+
+
+    printf("Segment text:\n%s", whisper_full_get_segment_text(wctx, 0));
+    assert(strcmp(" And so my fellow Americans, ask not what your country can do for you,"
+                  " ask what you can do for your country.",
+           whisper_full_get_segment_text(wctx, 0)) == 0);
+    assert(whisper_full_get_segment_t0(wctx, 0) == 32);
+    assert(whisper_full_get_segment_t1(wctx, 0) == 1051);
+
+    // token times mapped back to the original timeline: ordered, non-negative duration
+    const int n_tokens = whisper_full_n_tokens(wctx, 0);
+    assert(n_tokens > 0);
+    int64_t prev_t0 = -1;
+    for (int j = 0; j < n_tokens; ++j) {
+        const int64_t t0 = whisper_full_get_token_t0(wctx, 0, j);
+        const int64_t t1 = whisper_full_get_token_t1(wctx, 0, j);
+        assert(t0 >= 0 && t1 >= t0);
+        assert(t0 >= prev_t0);
+        prev_t0 = t0;
+    }
+
+    // internal VAD speech segments, on the original audio timeline (centiseconds)
+    const int n_vad = whisper_full_n_vad_segments(wctx);
+    assert(n_vad > 0);
+    int64_t vad_prev_end = -1;
+    for (int i = 0; i < n_vad; ++i) {
+        const int64_t t0 = whisper_full_get_vad_segment_t0(wctx, i);
+        const int64_t t1 = whisper_full_get_vad_segment_t1(wctx, i);
+        assert(t1 > t0);
+        assert(t0 >= vad_prev_end); // segments are ordered and non-overlapping
+        vad_prev_end = t1;
+    }
+
+    whisper_free(wctx);
+
+    return 0;
+}
--- a/tests/test-vad.cpp
+++ b/tests/test-vad.cpp
@@ -0,0 +1,85 @@
+#include "whisper.h"
+#include "common-whisper.h"
+
+#include <cstdio>
+#include <string>
+
+#ifdef NDEBUG
+#undef NDEBUG
+#endif
+#include <cassert>
+
+void assert_default_params(const struct whisper_vad_params & params) {
+    assert(params.threshold == 0.5);
+    assert(params.min_speech_duration_ms == 250);
+    assert(params.min_silence_duration_ms == 100);
+    assert(params.samples_overlap == 0.1f);
+}
+
+void assert_default_context_params(const struct whisper_vad_context_params & params) {
+    assert(params.n_threads == 4);
+    assert(params.use_gpu == false);
+    assert(params.gpu_device == 0);
+}
+
+void test_detect_speech(
+        struct whisper_vad_context * vctx,
+        struct whisper_vad_params params,
+        const float * pcmf32,
+        int n_samples) {
+    assert(whisper_vad_detect_speech(vctx, pcmf32, n_samples));
+    assert(whisper_vad_n_probs(vctx) == 344);
+    assert(whisper_vad_probs(vctx) != nullptr);
+}
+
+struct whisper_vad_segments * test_detect_timestamps(
+        struct whisper_vad_context * vctx,
+        struct whisper_vad_params params) {
+    struct whisper_vad_segments * timestamps = whisper_vad_segments_from_probs(vctx, params);
+    assert(whisper_vad_segments_n_segments(timestamps) == 4);
+
+    for (int i = 0; i < whisper_vad_segments_n_segments(timestamps); ++i) {
+        printf("VAD segment %d: start = %.2f, end = %.2f\n", i,
+               whisper_vad_segments_get_segment_t0(timestamps, i),
+               whisper_vad_segments_get_segment_t1(timestamps, i));
+    }
+
+    return timestamps;
+}
+
+int main() {
+    ggml_backend_load_all();
+
+    std::string vad_model_path = VAD_MODEL_PATH;
+    std::string sample_path    = SAMPLE_PATH;
+
+    // Load the sample audio file
+    std::vector<float> pcmf32;
+    std::vector<std::vector<float>> pcmf32s;
+    assert(read_audio_data(sample_path.c_str(), pcmf32, pcmf32s, false));
+    assert(pcmf32.size() > 0);
+    assert(pcmf32s.size() == 0); // no stereo vector
+
+    // Load the VAD model
+    struct whisper_vad_context_params ctx_params = whisper_vad_default_context_params();
+    assert_default_context_params(ctx_params);
+
+    struct whisper_vad_context * vctx = whisper_vad_init_from_file_with_params(
+            vad_model_path.c_str(),
+            ctx_params);
+    assert(vctx != nullptr);
+
+    struct whisper_vad_params params = whisper_vad_default_params();
+    assert_default_params(params);
+
+    // Test speech probabilites
+    test_detect_speech(vctx, params, pcmf32.data(), pcmf32.size());
+
+    // Test speech timestamps (uses speech probabilities from above)
+    struct whisper_vad_segments * timestamps = test_detect_timestamps(vctx, params);
+
+    whisper_vad_free_segments(timestamps);
+    whisper_vad_free(vctx);
+
+    return 0;
+}
--- a/tests/test-whisper-buffer-loader.cpp
+++ b/tests/test-whisper-buffer-loader.cpp
@@ -0,0 +1,24 @@
+#include "whisper.h"
+
+#include <cstdint>
+#include <cstdio>
+
+#ifdef NDEBUG
+#undef NDEBUG
+#endif
+#include <cassert>
+
+int main() {
+    struct whisper_context_params cparams = whisper_context_default_params();
+    cparams.use_gpu = false;
+
+    struct whisper_context * ctx_empty = whisper_init_from_buffer_with_params(nullptr, 1, cparams);
+    assert(ctx_empty == nullptr);
+
+    uint8_t truncated[8] = { 0 };
+    struct whisper_context * ctx_trunc = whisper_init_from_buffer_with_params(truncated, sizeof(truncated), cparams);
+    assert(ctx_trunc == nullptr);
+
+    printf("test-whisper-buffer-loader: OK\n");
+    return 0;
+}
--- a/tests/test-whisper-zero-samples.cpp
+++ b/tests/test-whisper-zero-samples.cpp
@@ -0,0 +1,32 @@
+#include "whisper.h"
+
+#include <cstdio>
+
+#ifdef NDEBUG
+#undef NDEBUG
+#endif
+#include <cassert>
+
+int main() {
+    ggml_backend_load_all();
+
+    struct whisper_context_params cparams = whisper_context_default_params();
+    cparams.use_gpu = false;
+
+    struct whisper_context * ctx = whisper_init_from_file_with_params(WHISPER_MODEL_PATH, cparams);
+    assert(ctx != nullptr);
+
+    struct whisper_full_params params = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
+    params.no_timestamps = true;
+    params.print_progress = false;
+    params.print_realtime = false;
+
+    const int rc = whisper_full(ctx, params, nullptr, 0);
+    assert(rc == 0);
+    assert(whisper_full_n_segments(ctx) == 0);
+
+    whisper_free(ctx);
+
+    printf("test-whisper-zero-samples: OK\n");
+    return 0;
+}
--- a/tests/test-whisper.js
+++ b/tests/test-whisper.js
@@ -0,0 +1,58 @@
+var factory = require('../bindings/javascript/whisper.js')
+
+factory().then(function(whisper) {
+    var fs = require('fs');
+
+    // to avoid reading WAV files and depending on some 3rd-party package, we read
+    // 32-bit float PCM directly. to genereate it:
+    //
+    //   $ ffmpeg -i samples/jfk.wav -f f32le -acodec pcm_f32le samples/jfk.pcmf32
+    //
+    let fname_wav   = "../samples/jfk.pcmf32";
+    let fname_model = "../models/ggml-base.en.bin";
+
+    // init whisper
+    {
+        // read binary data from file
+        var model_data = fs.readFileSync(fname_model);
+        if (model_data == null) {
+            console.log("whisper: failed to read model file");
+            process.exit(1);
+        }
+
+        // write binary data to WASM memory
+        whisper.FS_createDataFile("/", "whisper.bin", model_data, true, true);
+
+        // init the model
+        var ret = whisper.init("whisper.bin");
+        if (ret == false) {
+            console.log('whisper: failed to init');
+            process.exit(1);
+        }
+    }
+
+    // transcribe wav file
+    {
+        // read raw binary data
+        var pcm_data = fs.readFileSync(fname_wav);
+        if (pcm_data == null) {
+            console.log("whisper: failed to read wav file");
+            process.exit(1);
+        }
+
+        // convert to 32-bit float array
+        var pcm = new Float32Array(pcm_data.buffer);
+
+        // transcribe
+        var ret = whisper.full_default(pcm, "en", false);
+        if (ret != 0) {
+            console.log("whisper: failed to transcribe");
+            process.exit(1);
+        }
+    }
+
+    // free memory
+    {
+        whisper.free();
+    }
+});
				`@@ -0,0 +1 @@`
				My fellow Americans, this day has brought terrible news and great sadness to our country. At 9 o'clock this morning, Mission Control in Houston lost contact with our space shuttle, Columbia. A short time later, debris was seen falling from the skies above Texas. The Colombians lost. There are no survivors. On board was a crew of seven. Colonel Rick Husband, Lieutenant Colonel Michael Anderson, Commander Laurel Clark, Captain David Brown, Commander William McCool, Dr. Kultna Shavla, and Ilan Ramon, a colonel in the Israeli Air Force. These men and women assumed great risk in the service to all humanity. In an age when spaceflight has come to seem almost routine, it is easy to overlook the dangers of travel by rocket and the difficulties of navigating the fierce outer atmosphere of the Earth. These astronauts knew the dangers, and they faced them willingly, knowing they had a high and noble purpose in life. Because of their courage and daring and idealism, we will miss them all the more. All Americans today are thinking as well of the families of these men and women who have been given this sudden shock and grief. You're not alone. Our entire nation grieves with you. And those you love will always have the respect and gratitude of this country. The cause in which they died will continue. Mankind is led into the darkness beyond our world by the inspiration of discovery and the longing to understand. Our journey into space will go on. In the skies today, we saw destruction and tragedy. Yet farther than we can see, there is comfort and hope. In the words of the prophet Isaiah, "Lift your eyes and look to the heavens. Who created all these? He who brings out the starry hosts one by one and calls them each by name." Because of His great power and mighty strength, not one of them is missing. The same Creator who names the stars also knows the names of the seven souls we mourn today. The crew of the shuttle Columbia did not return safely to Earth, yet we can pray that all are safely home. May God bless the grieving families. And may God continue to bless America. [Silence]
				`@@ -0,0 +1 @@`
				Henry F. Phillips from Wikipedia, the free encyclopedia at en.wikipedia.org. Henry F. Phillips from Wikipedia, the free encyclopedia. Henry F. Phillips 1890-1958, a U.S. businessman from Portland, Oregon, has the honor of having the Phillips head screw and screwdriver named after him. The importance of the cross head screw design lies in its self-centering property, useful on automated production lines that use powered screwdrivers. Phillips' major contribution was in driving the cross head concept forward to the point where it was adopted by screw makers and automobile companies. Although he received patents for the design in 1936, U.S. Patent #2,046,343, U.S. Patents #2,046,837 to #2,046,840, it was so widely copied that by 1949 Phillips lost his patent. The American Screw Company was responsible for devising a means of manufacturing the screw, and successfully patented and licensed their method. Other screw makers of the 1930s dismissed the Phillips concept since it calls for a relatively complex recessed socket shape in the head of the screw, as distinct from the simple milled slot of a slotted type screw. The Phillips Screw Company and the American Screw Company went on to devise the Pawsadrive screw, which differs from the Phillips in that it is designed to accommodate greater torque than the Phillips. An image accompanied this article, captioned "Phillips Screw Head." The following is an info box which accompanies this article. Info box, part of the series on screw drive types. Slotted, commonly erroneously flat head. Phillips, cross head. Pawsadrive, super drive. Torques. Hex, Allen. Robertson. Tri-wing. Torx set. Spanner head. Triple square, XZN. Others, poly drive, spline drive, double hex. Many images accompanied this info box. This page was last modified on the 9th of April, 2008, at 1704. All text is available under the terms of the GNU Free Documentation License. See copyrights for details. Wikipedia is a registered trademark of the Wikimedia Foundation Incorporated, a U.S. registered 501(c)(3) tax-deductible nonprofit charity. This sound file and all text in the article are licensed under the GNU Free Documentation License, available at www.gnu.org/copyleft/fdl.html.
				`@@ -0,0 +1 @@`
				This is the Micro Machine Man presenting the most midget miniature motorcade of Micro Machines. Each one has dramatic details, terrific trim, precision paint jobs, plus incredible Micro Machine Pocket Playsets. There's a police station, fire station, restaurant, service station, and more. Perfect pocket portables to take anyplace. And there are many miniature playsets to play with, and each one comes with its own special edition Micro Machine vehicle and fun, fantastic features that miraculously move. Raise the boat lift at the airport marina, man the gun turret at the army base, clean your car at the car wash, raise the toll bridge. And these playsets fit together to form a Micro Machine world. Micro Machine Pocket Playsets, so tremendously tiny, so perfectly precise, so dazzlingly detailed, you'll want to pocket them all. Micro Machines are Micro Machine Pocket Playsets sold separately from Galoob. The smaller they are, the better they are.
				`@@ -0,0 +1 @@`
				Hola, como están todos? Mi nombre es Julián Virrueta Mendoza y en este podcast les vengo a hablar sobre la contaminación del agua. Bueno, empezaré por decir que el ser humano no está midiendo las consecuencias de sus actos. No hay duda que uno de los mayores problemas a los que se enfrentan muchas poblaciones actualmente es la contaminación del agua. Principalmente porque como bien sabemos el agua prácticamente es fundamental para la vida, por lo que la contaminación puede ser algo muy negativo para el desarrollo tanto económico como social de los pueblos o de las poblaciones próximas en ese lugar contaminado. Los comienzos de la contaminación, como lo definen muchos expertos en la materia, la contaminación del agua es causada por las actividades humanas. Es un fenómeno ambiental de importancia, el cual se comienza a producir desde los primeros intentos de industrialización para transformarse luego en un problema tan habitual como generalizado. Generalmente la contaminación del agua se produce a través de la introducción directa o indirecta en los acuíferos o caos de agua, ríos, mares, lagos, océanos, etc. o de diversas sustancias que pueden ser consideradas como contaminantes. Pero existen dos formas principales de contaminación del agua. Una de ellas tiene que ver con la contaminación natural del agua que se corresponde con el ciclo natural de esta durante el que puede entrar en contacto con ciertos constituyentes contaminantes como sustancias minerales y orgánicas disueltas o en suspensión que se vierten en la corteza terrestre, la atmósfera y en las aguas. Pero todo esto se puede contradecir si el ser humano comía sus consecuencias, si no tirara basura a los lagos, a los ríos, no tirara botes de aceite, no contaminara. Bueno amigos, yo los invito a que no contaminen el agua y que sepan cuidar la naturaleza. Los saluda su buen amigo y compañero Julián Virreta. Nos vemos. ¡Claro!
				`@@ -0,0 +1 @@`
				`And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.`