make faster_whisper.assets as a valid python package to distribute (#772) (#774)

* Foolproof: Disable VAD if clip_timestamps is in use

Prevent silly things to happen.

CONTRIBUTING.md

@@ -7,7 +7,7 @@ Contributions are welcome! Here are some pointers to help you install the librar
 We recommend installing the module in editable mode with the `dev` extra requirements:
 
 ```bash
-git clone https://github.com/guillaumekln/faster-whisper.git
+git clone https://github.com/SYSTRAN/faster-whisper.git
 cd faster-whisper/
 pip install -e .[dev]
 ```

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2023 Guillaume Klein
+Copyright (c) 2023 SYSTRAN
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -1,4 +1,4 @@
-[![CI](https://github.com/guillaumekln/faster-whisper/workflows/CI/badge.svg)](https://github.com/guillaumekln/faster-whisper/actions?query=workflow%3ACI) [![PyPI version](https://badge.fury.io/py/faster-whisper.svg)](https://badge.fury.io/py/faster-whisper)
+[![CI](https://github.com/SYSTRAN/faster-whisper/workflows/CI/badge.svg)](https://github.com/SYSTRAN/faster-whisper/actions?query=workflow%3ACI) [![PyPI version](https://badge.fury.io/py/faster-whisper.svg)](https://badge.fury.io/py/faster-whisper)
 
 # Faster Whisper transcription with CTranslate2
 
@@ -8,11 +8,13 @@ This implementation is up to 4 times faster than [openai/whisper](https://github
 
 ## Benchmark
 
+### Whisper
+
 For reference, here's the time and memory usage that are required to transcribe [**13 minutes**](https://www.youtube.com/watch?v=0u7tTptBo9I) of audio using different implementations:
 
 * [openai/whisper](https://github.com/openai/whisper)@[6dea21fd](https://github.com/openai/whisper/commit/6dea21fd7f7253bfe450f1e2512a0fe47ee2d258)
 * [whisper.cpp](https://github.com/ggerganov/whisper.cpp)@[3b010f9](https://github.com/ggerganov/whisper.cpp/commit/3b010f9bed9a6068609e9faf52383aea792b0362)
-* [faster-whisper](https://github.com/guillaumekln/faster-whisper)@[cce6b53e](https://github.com/guillaumekln/faster-whisper/commit/cce6b53e4554f71172dad188c45f10fb100f6e3e)
+* [faster-whisper](https://github.com/SYSTRAN/faster-whisper)@[cce6b53e](https://github.com/SYSTRAN/faster-whisper/commit/cce6b53e4554f71172dad188c45f10fb100f6e3e)
 
 ### Large-v2 model on GPU
 
@@ -36,6 +38,71 @@ For reference, here's the time and memory usage that are required to transcribe
 
 *Executed with 8 threads on a Intel(R) Xeon(R) Gold 6226R.*
 
+
+### Distil-whisper
+
+| Implementation | Precision | Beam size | Time | Gigaspeech WER |
+| --- | --- | --- | --- | --- |
+| distil-whisper/distil-large-v2 | fp16 | 4 |- | 10.36 |
+| [faster-distil-large-v2](https://huggingface.co/Systran/faster-distil-whisper-large-v2) | fp16 | 5 | - | 10.28 |
+| distil-whisper/distil-medium.en | fp16 | 4 | - | 11.21 |
+| [faster-distil-medium.en](https://huggingface.co/Systran/faster-distil-whisper-medium.en) | fp16 | 5 | - | 11.21 |
+
+*Executed with CUDA 11.4 on a NVIDIA 3090.*
+
+<details>
+<summary>testing details (click to expand)</summary>
+
+For `distil-whisper/distil-large-v2`, the WER is tested with code sample from [link](https://huggingface.co/distil-whisper/distil-large-v2#evaluation). for `faster-distil-whisper`, the WER is tested with setting:
+```python
+from faster_whisper import WhisperModel
+
+model_size = "distil-large-v2"
+# model_size = "distil-medium.en"
+# Run on GPU with FP16
+model = WhisperModel(model_size, device="cuda", compute_type="float16")
+segments, info = model.transcribe("audio.mp3", beam_size=5, language="en")
+```
+</details>
+
+## Requirements
+
+* Python 3.8 or greater
+
+Unlike openai-whisper, FFmpeg does **not** need to be installed on the system. The audio is decoded with the Python library [PyAV](https://github.com/PyAV-Org/PyAV) which bundles the FFmpeg libraries in its package.
+
+### GPU
+
+GPU execution requires the following NVIDIA libraries to be installed:
+
+* [cuBLAS for CUDA 11](https://developer.nvidia.com/cublas)
+* [cuDNN 8 for CUDA 11](https://developer.nvidia.com/cudnn)
+
+There are multiple ways to install these libraries. The recommended way is described in the official NVIDIA documentation, but we also suggest other installation methods below.
+
+<details>
+<summary>Other installation methods (click to expand)</summary>
+
+#### Use Docker
+
+The libraries are installed in this official NVIDIA Docker image: `nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04`.
+
+#### Install with `pip` (Linux only)
+
+On Linux these libraries can be installed with `pip`. Note that `LD_LIBRARY_PATH` must be set before launching Python.
+
+```bash
+pip install nvidia-cublas-cu11 nvidia-cudnn-cu11
+
+export LD_LIBRARY_PATH=`python3 -c 'import os; import nvidia.cublas.lib; import nvidia.cudnn.lib; print(os.path.dirname(nvidia.cublas.lib.__file__) + ":" + os.path.dirname(nvidia.cudnn.lib.__file__))'`
+```
+
+#### Download the libraries from Purfview's repository (Windows & Linux)
+
+Purfview's [whisper-standalone-win](https://github.com/Purfview/whisper-standalone-win) provides the required NVIDIA libraries for Windows & Linux in a [single archive](https://github.com/Purfview/whisper-standalone-win/releases/tag/libs). Decompress the archive and place the libraries in a directory included in the `PATH`.
+
+</details>
+
 ## Installation
 
 The module can be installed from [PyPI](https://pypi.org/project/faster-whisper/):
@@ -44,26 +111,31 @@ The module can be installed from [PyPI](https://pypi.org/project/faster-whisper/
 pip install faster-whisper
 ```
 
-**Other installation methods:**
+<details>
+<summary>Other installation methods (click to expand)</summary>
+
+### Install the master branch
 
 ```bash
-# Install the master branch:
-pip install --force-reinstall "faster-whisper @ https://github.com/guillaumekln/faster-whisper/archive/refs/heads/master.tar.gz"
-
-# Install a specific commit:
-pip install --force-reinstall "faster-whisper @ https://github.com/guillaumekln/faster-whisper/archive/a4f1cc8f11433e454c3934442b5e1a4ed5e865c3.tar.gz"
+pip install --force-reinstall "faster-whisper @ https://github.com/SYSTRAN/faster-whisper/archive/refs/heads/master.tar.gz"
 ```
 
-### GPU support
+### Install a specific commit
 
-GPU execution requires the NVIDIA libraries cuBLAS 11.x and cuDNN 8.x to be installed on the system. Please refer to the [CTranslate2 documentation](https://opennmt.net/CTranslate2/installation.html).
+```bash
+pip install --force-reinstall "faster-whisper @ https://github.com/SYSTRAN/faster-whisper/archive/a4f1cc8f11433e454c3934442b5e1a4ed5e865c3.tar.gz"
+```
+
+</details>
 
 ## Usage
 
+### Faster-whisper
+
 ```python
 from faster_whisper import WhisperModel
 
-model_size = "large-v2"
+model_size = "large-v3"
 
 # Run on GPU with FP16
 model = WhisperModel(model_size, device="cuda", compute_type="float16")
@@ -87,6 +159,25 @@ for segment in segments:
 segments, _ = model.transcribe("audio.mp3")
 segments = list(segments)  # The transcription will actually run here.
 ```
+### Faster Distil-Whisper
+
+The Distil-Whisper checkpoints are compatible with the Faster-Whisper package. In particular, the latest [distil-large-v3](https://huggingface.co/distil-whisper/distil-large-v3)
+checkpoint is intrinsically designed to work with the Faster-Whisper transcription algorithm. The following code snippet 
+demonstrates how to run inference with distil-large-v3 on a specified audio file:
+
+```python
+from faster_whisper import WhisperModel
+
+model_size = "distil-large-v3"
+
+model = WhisperModel(model_size, device="cuda", compute_type="float16")
+segments, info = model.transcribe("audio.mp3", beam_size=5, language="en", condition_on_previous_text=False)
+
+for segment in segments:
+    print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))
+```
+
+For more information about the distil-large-v3 model, refer to the original [model card](https://huggingface.co/distil-whisper/distil-large-v3).
 
 ### Word-level timestamps
 
@@ -106,7 +197,7 @@ The library integrates the [Silero VAD](https://github.com/snakers4/silero-vad)
 segments, _ = model.transcribe("audio.mp3", vad_filter=True)
 ```
 
-The default behavior is conservative and only removes silence longer than 2 seconds. See the available VAD parameters and default values in the [source code](https://github.com/guillaumekln/faster-whisper/blob/master/faster_whisper/vad.py). They can be customized with the dictionary argument `vad_parameters`:
+The default behavior is conservative and only removes silence longer than 2 seconds. See the available VAD parameters and default values in the [source code](https://github.com/SYSTRAN/faster-whisper/blob/master/faster_whisper/vad.py). They can be customized with the dictionary argument `vad_parameters`:
 
 ```python
 segments, _ = model.transcribe(
@@ -129,31 +220,38 @@ logging.getLogger("faster_whisper").setLevel(logging.DEBUG)
 
 ### Going further
 
-See more model and transcription options in the [`WhisperModel`](https://github.com/guillaumekln/faster-whisper/blob/master/faster_whisper/transcribe.py) class implementation.
+See more model and transcription options in the [`WhisperModel`](https://github.com/SYSTRAN/faster-whisper/blob/master/faster_whisper/transcribe.py) class implementation.
 
 ## Community integrations
 
 Here is a non exhaustive list of open-source projects using faster-whisper. Feel free to add your project to the list!
 
+
+* [WhisperX](https://github.com/m-bain/whisperX) is an award-winning Python library that offers speaker diarization and accurate word-level timestamps using wav2vec2 alignment
 * [whisper-ctranslate2](https://github.com/Softcatala/whisper-ctranslate2) is a command line client based on faster-whisper and compatible with the original client from openai/whisper.
 * [whisper-diarize](https://github.com/MahmoudAshraf97/whisper-diarization) is a speaker diarization tool that is based on faster-whisper and NVIDIA NeMo.
-* [whisper-standalone-win](https://github.com/Purfview/whisper-standalone-win) contains the portable ready to run binaries of faster-whisper for Windows.
+* [whisper-standalone-win](https://github.com/Purfview/whisper-standalone-win) Standalone CLI executables of faster-whisper for Windows, Linux & macOS. 
 * [asr-sd-pipeline](https://github.com/hedrergudene/asr-sd-pipeline) provides a scalable, modular, end to end multi-speaker speech to text solution implemented using AzureML pipelines.
 * [Open-Lyrics](https://github.com/zh-plus/Open-Lyrics) is a Python library that transcribes voice files using faster-whisper, and translates/polishes the resulting text into `.lrc` files in the desired language using OpenAI-GPT.
+* [wscribe](https://github.com/geekodour/wscribe) is a flexible transcript generation tool supporting faster-whisper, it can export word level transcript and the exported transcript then can be edited with [wscribe-editor](https://github.com/geekodour/wscribe-editor)
+* [aTrain](https://github.com/BANDAS-Center/aTrain) is a graphical user interface implementation of faster-whisper developed at the BANDAS-Center at the University of Graz for transcription and diarization in Windows ([Windows Store App](https://apps.microsoft.com/detail/atrain/9N15Q44SZNS2)) and Linux.
+* [Whisper-Streaming](https://github.com/ufal/whisper_streaming) implements real-time mode for offline Whisper-like speech-to-text models with faster-whisper as the most recommended back-end. It implements a streaming policy with self-adaptive latency based on the actual source complexity, and demonstrates the state of the art.
+* [WhisperLive](https://github.com/collabora/WhisperLive) is a nearly-live implementation of OpenAI's Whisper which uses faster-whisper as the backend to transcribe audio in real-time.
+* [Faster-Whisper-Transcriber](https://github.com/BBC-Esq/ctranslate2-faster-whisper-transcriber) is a simple but reliable voice transcriber that provides a user-friendly interface.
 
 ## Model conversion
 
-When loading a model from its size such as `WhisperModel("large-v2")`, the correspondig CTranslate2 model is automatically downloaded from the [Hugging Face Hub](https://huggingface.co/guillaumekln).
+When loading a model from its size such as `WhisperModel("large-v3")`, the corresponding CTranslate2 model is automatically downloaded from the [Hugging Face Hub](https://huggingface.co/Systran).
 
 We also provide a script to convert any Whisper models compatible with the Transformers library. They could be the original OpenAI models or user fine-tuned models.
 
-For example the command below converts the [original "large-v2" Whisper model](https://huggingface.co/openai/whisper-large-v2) and saves the weights in FP16:
+For example the command below converts the [original "large-v3" Whisper model](https://huggingface.co/openai/whisper-large-v3) and saves the weights in FP16:
 
 ```bash
 pip install transformers[torch]>=4.23
 
-ct2-transformers-converter --model openai/whisper-large-v2 --output_dir whisper-large-v2-ct2 \
-    --copy_files tokenizer.json --quantization float16
+ct2-transformers-converter --model openai/whisper-large-v3 --output_dir whisper-large-v3-ct2
+--copy_files tokenizer.json preprocessor_config.json --quantization float16
 ```
 
 * The option `--model` accepts a model name on the Hub or a path to a model directory.
@@ -165,12 +263,12 @@ Models can also be converted from the code. See the [conversion API](https://ope
 
 1. Directly load the model from a local directory:
 ```python
-model = faster_whisper.WhisperModel('whisper-large-v2-ct2')
+model = faster_whisper.WhisperModel("whisper-large-v3-ct2")
 ```
 
 2. [Upload your model to the Hugging Face Hub](https://huggingface.co/docs/transformers/model_sharing#upload-with-the-web-interface) and load it from its name:
 ```python
-model = faster_whisper.WhisperModel('username/whisper-large-v2-ct2')
+model = faster_whisper.WhisperModel("username/whisper-large-v3-ct2")
 ```
 
 ## Comparing performance against other implementations

faster_whisper/__init__.py

@@ -1,9 +1,10 @@
 from faster_whisper.audio import decode_audio
 from faster_whisper.transcribe import WhisperModel
-from faster_whisper.utils import download_model, format_timestamp
+from faster_whisper.utils import available_models, download_model, format_timestamp
 from faster_whisper.version import __version__
 
 __all__ = [
+    "available_models",
     "decode_audio",
     "WhisperModel",
     "download_model",

faster_whisper/audio.py

@@ -6,6 +6,7 @@ system dependencies. FFmpeg does not need to be installed on the system.
 However, the API is quite low-level so we need to manipulate audio frames directly.
 """
 
+import gc
 import io
 import itertools
 
@@ -42,7 +43,7 @@ def decode_audio(
     raw_buffer = io.BytesIO()
     dtype = None
 
-    with av.open(input_file, metadata_errors="ignore") as container:
+    with av.open(input_file, mode="r", metadata_errors="ignore") as container:
         frames = container.decode(audio=0)
         frames = _ignore_invalid_frames(frames)
         frames = _group_frames(frames, 500000)
@@ -53,6 +54,11 @@ def decode_audio(
             dtype = array.dtype
             raw_buffer.write(array)
 
+    # It appears that some objects related to the resampler are not freed
+    # unless the garbage collector is manually run.
+    del resampler
+    gc.collect()
+
     audio = np.frombuffer(raw_buffer.getbuffer(), dtype=dtype)
 
     # Convert s16 back to f32.
@@ -96,3 +102,18 @@ def _resample_frames(frames, resampler):
     # Add None to flush the resampler.
     for frame in itertools.chain(frames, [None]):
         yield from resampler.resample(frame)
+
+
+def pad_or_trim(array, length: int, *, axis: int = -1):
+    """
+    Pad or trim the audio array to N_SAMPLES, as expected by the encoder.
+    """
+    if array.shape[axis] > length:
+        array = array.take(indices=range(length), axis=axis)
+
+    if array.shape[axis] < length:
+        pad_widths = [(0, 0)] * array.ndim
+        pad_widths[axis] = (0, length - array.shape[axis])
+        array = np.pad(array, pad_widths)
+
+    return array

faster_whisper/feature_extractor.py

@@ -142,11 +142,15 @@ class FeatureExtractor:
             data[f] = np.fft.fft(fft_signal, axis=0)[:num_fft_bins]
         return data.T
 
-    def __call__(self, waveform, padding=True):
+    def __call__(self, waveform, padding=True, chunk_length=None):
         """
         Compute the log-Mel spectrogram of the provided audio, gives similar results
         whisper's original torch implementation with 1e-5 tolerance.
         """
+        if chunk_length is not None:
+            self.n_samples = chunk_length * self.sampling_rate
+            self.nb_max_frames = self.n_samples // self.hop_length
+
         if padding:
             waveform = np.pad(waveform, [(0, self.n_samples)])
 

faster_whisper/tokenizer.py

@@ -19,15 +19,21 @@ class Tokenizer:
         self.tokenizer = tokenizer
 
         if multilingual:
+            if task not in _TASKS:
+                raise ValueError(
+                    "'%s' is not a valid task (accepted tasks: %s)"
+                    % (task, ", ".join(_TASKS))
+                )
+
+            if language not in _LANGUAGE_CODES:
+                raise ValueError(
+                    "'%s' is not a valid language code (accepted language codes: %s)"
+                    % (language, ", ".join(_LANGUAGE_CODES))
+                )
+
             self.task = self.tokenizer.token_to_id("<|%s|>" % task)
-            if self.task is None:
-                raise ValueError("%s is not a valid task" % task)
-
-            self.language_code = language
             self.language = self.tokenizer.token_to_id("<|%s|>" % language)
-            if self.language is None:
-                raise ValueError("%s is not a valid language code" % language)
-
+            self.language_code = language
         else:
             self.task = None
             self.language = None
@@ -102,7 +108,7 @@ class Tokenizer:
     def split_to_word_tokens(
         self, tokens: List[int]
     ) -> Tuple[List[str], List[List[int]]]:
-        if self.language_code in {"zh", "ja", "th", "lo", "my"}:
+        if self.language_code in {"zh", "ja", "th", "lo", "my", "yue"}:
             # These languages don't typically use spaces, so it is difficult to split words
             # without morpheme analysis. Here, we instead split words at any
             # position where the tokens are decoded as valid unicode points
@@ -161,3 +167,112 @@ class Tokenizer:
                 word_tokens[-1].extend(subword_tokens)
 
         return words, word_tokens
+
+
+_TASKS = (
+    "transcribe",
+    "translate",
+)
+
+_LANGUAGE_CODES = (
+    "af",
+    "am",
+    "ar",
+    "as",
+    "az",
+    "ba",
+    "be",
+    "bg",
+    "bn",
+    "bo",
+    "br",
+    "bs",
+    "ca",
+    "cs",
+    "cy",
+    "da",
+    "de",
+    "el",
+    "en",
+    "es",
+    "et",
+    "eu",
+    "fa",
+    "fi",
+    "fo",
+    "fr",
+    "gl",
+    "gu",
+    "ha",
+    "haw",
+    "he",
+    "hi",
+    "hr",
+    "ht",
+    "hu",
+    "hy",
+    "id",
+    "is",
+    "it",
+    "ja",
+    "jw",
+    "ka",
+    "kk",
+    "km",
+    "kn",
+    "ko",
+    "la",
+    "lb",
+    "ln",
+    "lo",
+    "lt",
+    "lv",
+    "mg",
+    "mi",
+    "mk",
+    "ml",
+    "mn",
+    "mr",
+    "ms",
+    "mt",
+    "my",
+    "ne",
+    "nl",
+    "nn",
+    "no",
+    "oc",
+    "pa",
+    "pl",
+    "ps",
+    "pt",
+    "ro",
+    "ru",
+    "sa",
+    "sd",
+    "si",
+    "sk",
+    "sl",
+    "sn",
+    "so",
+    "sq",
+    "sr",
+    "su",
+    "sv",
+    "sw",
+    "ta",
+    "te",
+    "tg",
+    "th",
+    "tk",
+    "tl",
+    "tr",
+    "tt",
+    "uk",
+    "ur",
+    "uz",
+    "vi",
+    "yi",
+    "yo",
+    "zh",
+    "yue",
+)

faster_whisper/transcribe.py

@@ -1,18 +1,20 @@
 import itertools
+import json
 import logging
 import os
 import zlib
 
+from inspect import signature
 from typing import BinaryIO, Iterable, List, NamedTuple, Optional, Tuple, Union
 
 import ctranslate2
 import numpy as np
 import tokenizers
 
-from faster_whisper.audio import decode_audio
+from faster_whisper.audio import decode_audio, pad_or_trim
 from faster_whisper.feature_extractor import FeatureExtractor
-from faster_whisper.tokenizer import Tokenizer
-from faster_whisper.utils import download_model, format_timestamp, get_logger
+from faster_whisper.tokenizer import _LANGUAGE_CODES, Tokenizer
+from faster_whisper.utils import download_model, format_timestamp, get_end, get_logger
 from faster_whisper.vad import (
     SpeechTimestampsMap,
     VadOptions,
@@ -47,10 +49,13 @@ class TranscriptionOptions(NamedTuple):
     best_of: int
     patience: float
     length_penalty: float
+    repetition_penalty: float
+    no_repeat_ngram_size: int
     log_prob_threshold: Optional[float]
     no_speech_threshold: Optional[float]
     compression_ratio_threshold: Optional[float]
     condition_on_previous_text: bool
+    prompt_reset_on_temperature: float
     temperatures: List[float]
     initial_prompt: Optional[Union[str, Iterable[int]]]
     prefix: Optional[str]
@@ -61,12 +66,16 @@ class TranscriptionOptions(NamedTuple):
     word_timestamps: bool
     prepend_punctuations: str
     append_punctuations: str
+    max_new_tokens: Optional[int]
+    clip_timestamps: Union[str, List[float]]
+    hallucination_silence_threshold: Optional[float]
 
 
 class TranscriptionInfo(NamedTuple):
     language: str
     language_probability: float
     duration: float
+    duration_after_vad: float
     all_language_probs: Optional[List[Tuple[str, float]]]
     transcription_options: TranscriptionOptions
     vad_options: VadOptions
@@ -88,8 +97,8 @@ class WhisperModel:
 
         Args:
           model_size_or_path: Size of the model to use (tiny, tiny.en, base, base.en,
-            small, small.en, medium, medium.en, large-v1, or large-v2), a path to a converted
-            model directory, or a CTranslate2-converted Whisper model ID from the Hugging Face Hub.
+            small, small.en, medium, medium.en, large-v1, large-v2, large-v3, or large), a path to a
+            converted model directory, or a CTranslate2-converted Whisper model ID from the HF Hub.
             When a size or a model ID is configured, the converted model is downloaded
             from the Hugging Face Hub.
           device: Device to use for computation ("cpu", "cuda", "auto").
@@ -138,7 +147,8 @@ class WhisperModel:
                 "openai/whisper-tiny" + ("" if self.model.is_multilingual else ".en")
             )
 
-        self.feature_extractor = FeatureExtractor()
+        self.feat_kwargs = self._get_feature_kwargs(model_path)
+        self.feature_extractor = FeatureExtractor(**self.feat_kwargs)
         self.num_samples_per_token = self.feature_extractor.hop_length * 2
         self.frames_per_second = (
             self.feature_extractor.sampling_rate // self.feature_extractor.hop_length
@@ -150,6 +160,27 @@ class WhisperModel:
         self.time_precision = 0.02
         self.max_length = 448
 
+    @property
+    def supported_languages(self) -> List[str]:
+        """The languages supported by the model."""
+        return list(_LANGUAGE_CODES) if self.model.is_multilingual else ["en"]
+
+    def _get_feature_kwargs(self, model_path) -> dict:
+        preprocessor_config_file = os.path.join(model_path, "preprocessor_config.json")
+        config = {}
+        if os.path.isfile(preprocessor_config_file):
+            try:
+                with open(preprocessor_config_file, "r", encoding="utf-8") as json_file:
+                    config = json.load(json_file)
+                valid_keys = signature(FeatureExtractor.__init__).parameters.keys()
+                config = {k: v for k, v in config.items() if k in valid_keys}
+            except json.JSONDecodeError as e:
+                self.logger.warning(
+                    "Could not load preprocessor_config.json: %s", str(e)
+                )
+
+        return config
+
     def transcribe(
         self,
         audio: Union[str, BinaryIO, np.ndarray],
@@ -159,6 +190,8 @@ class WhisperModel:
         best_of: int = 5,
         patience: float = 1,
         length_penalty: float = 1,
+        repetition_penalty: float = 1,
+        no_repeat_ngram_size: int = 0,
         temperature: Union[float, List[float], Tuple[float, ...]] = [
             0.0,
             0.2,
@@ -171,6 +204,7 @@ class WhisperModel:
         log_prob_threshold: Optional[float] = -1.0,
         no_speech_threshold: Optional[float] = 0.6,
         condition_on_previous_text: bool = True,
+        prompt_reset_on_temperature: float = 0.5,
         initial_prompt: Optional[Union[str, Iterable[int]]] = None,
         prefix: Optional[str] = None,
         suppress_blank: bool = True,
@@ -182,6 +216,12 @@ class WhisperModel:
         append_punctuations: str = "\"'.。,，!！?？:：”)]}、",
         vad_filter: bool = False,
         vad_parameters: Optional[Union[dict, VadOptions]] = None,
+        max_new_tokens: Optional[int] = None,
+        chunk_length: Optional[int] = None,
+        clip_timestamps: Union[str, List[float]] = "0",
+        hallucination_silence_threshold: Optional[float] = None,
+        language_detection_threshold: Optional[float] = None,
+        language_detection_segments: int = 1,
     ) -> Tuple[Iterable[Segment], TranscriptionInfo]:
         """Transcribes an input file.
 
@@ -195,6 +235,9 @@ class WhisperModel:
           best_of: Number of candidates when sampling with non-zero temperature.
           patience: Beam search patience factor.
           length_penalty: Exponential length penalty constant.
+          repetition_penalty: Penalty applied to the score of previously generated tokens
+            (set > 1 to penalize).
+          no_repeat_ngram_size: Prevent repetitions of ngrams with this size (set 0 to disable).
           temperature: Temperature for sampling. It can be a tuple of temperatures,
             which will be successively used upon failures according to either
             `compression_ratio_threshold` or `log_prob_threshold`.
@@ -209,6 +252,8 @@ class WhisperModel:
             as a prompt for the next window; disabling may make the text inconsistent across
             windows, but the model becomes less prone to getting stuck in a failure loop,
             such as repetition looping or timestamps going out of sync.
+          prompt_reset_on_temperature: Resets prompt if temperature is above this value.
+            Arg has effect only if condition_on_previous_text is True.
           initial_prompt: Optional text string or iterable of token ids to provide as a
             prompt for the first window.
           prefix: Optional text to provide as a prefix for the first window.
@@ -228,6 +273,20 @@ class WhisperModel:
             https://github.com/snakers4/silero-vad.
           vad_parameters: Dictionary of Silero VAD parameters or VadOptions class (see available
             parameters and default values in the class `VadOptions`).
+          max_new_tokens: Maximum number of new tokens to generate per-chunk. If not set,
+            the maximum will be set by the default max_length.
+          chunk_length: The length of audio segments. If it is not None, it will overwrite the
+            default chunk_length of the FeatureExtractor.
+          clip_timestamps: Union[str, List[float]]
+            Comma-separated list start,end,start,end,... timestamps (in seconds) of clips to
+             process. The last end timestamp defaults to the end of the file.
+             vad_filter will be ignored if clip_timestamps is used.
+          hallucination_silence_threshold: Optional[float]
+            When word_timestamps is True, skip silent periods longer than this threshold
+             (in seconds) when a possible hallucination is detected
+          language_detection_threshold: If the maximum probability of the language tokens is higher
+           than this value, the language is detected.
+          language_detection_segments: Number of segments to consider for the language detection.
 
         Returns:
           A tuple with:
@@ -241,22 +300,24 @@ class WhisperModel:
             audio = decode_audio(audio, sampling_rate=sampling_rate)
 
         duration = audio.shape[0] / sampling_rate
+        duration_after_vad = duration
 
         self.logger.info(
             "Processing audio with duration %s", format_timestamp(duration)
         )
 
-        if vad_filter:
+        if vad_filter and clip_timestamps == "0":
             if vad_parameters is None:
                 vad_parameters = VadOptions()
             elif isinstance(vad_parameters, dict):
                 vad_parameters = VadOptions(**vad_parameters)
             speech_chunks = get_speech_timestamps(audio, vad_parameters)
             audio = collect_chunks(audio, speech_chunks)
+            duration_after_vad = audio.shape[0] / sampling_rate
 
             self.logger.info(
                 "VAD filter removed %s of audio",
-                format_timestamp(duration - (audio.shape[0] / sampling_rate)),
+                format_timestamp(duration - duration_after_vad),
             )
 
             if self.logger.isEnabledFor(logging.DEBUG):
@@ -275,7 +336,7 @@ class WhisperModel:
         else:
             speech_chunks = None
 
-        features = self.feature_extractor(audio)
+        features = self.feature_extractor(audio, chunk_length=chunk_length)
 
         encoder_output = None
         all_language_probs = None
@@ -285,15 +346,51 @@ class WhisperModel:
                 language = "en"
                 language_probability = 1
             else:
-                segment = features[:, : self.feature_extractor.nb_max_frames]
-                encoder_output = self.encode(segment)
-                # results is a list of tuple[str, float] with language names and
-                # probabilities.
-                results = self.model.detect_language(encoder_output)[0]
-                # Parse language names to strip out markers
-                all_language_probs = [(token[2:-2], prob) for (token, prob) in results]
-                # Get top language token and probability
-                language, language_probability = all_language_probs[0]
+                if (
+                    language_detection_segments is None
+                    or language_detection_segments < 1
+                ):
+                    language_detection_segments = 1
+                seek = 0
+                detected_language_info = {}
+                content_frames = (
+                    features.shape[-1] - self.feature_extractor.nb_max_frames
+                )
+                while (
+                    seek <= content_frames
+                    and seek
+                    < self.feature_extractor.nb_max_frames * language_detection_segments
+                ):
+                    segment = features[
+                        :, seek : seek + self.feature_extractor.nb_max_frames
+                    ]
+                    encoder_output = self.encode(segment)
+                    # results is a list of tuple[str, float] with language names and
+                    # probabilities.
+                    results = self.model.detect_language(encoder_output)[0]
+                    # Parse language names to strip out markers
+                    all_language_probs = [
+                        (token[2:-2], prob) for (token, prob) in results
+                    ]
+                    # Get top language token and probability
+                    language, language_probability = all_language_probs[0]
+                    if (
+                        language_detection_threshold is None
+                        or language_probability > language_detection_threshold
+                    ):
+                        break
+                    detected_language_info.setdefault(language, []).append(
+                        language_probability
+                    )
+                    seek += segment.shape[-1]
+                else:
+                    # If no language detected for all segments, the majority vote of the highest
+                    # projected languages for all segments is used to determine the language.
+                    language = max(
+                        detected_language_info,
+                        key=lambda lang: len(detected_language_info[lang]),
+                    )
+                    language_probability = max(detected_language_info[language])
 
                 self.logger.info(
                     "Detected language '%s' with probability %.2f",
@@ -301,6 +398,13 @@ class WhisperModel:
                     language_probability,
                 )
         else:
+            if not self.model.is_multilingual and language != "en":
+                self.logger.warning(
+                    "The current model is English-only but the language parameter is set to '%s'; "
+                    "using 'en' instead." % language
+                )
+                language = "en"
+
             language_probability = 1
 
         tokenizer = Tokenizer(
@@ -315,10 +419,13 @@ class WhisperModel:
             best_of=best_of,
             patience=patience,
             length_penalty=length_penalty,
+            repetition_penalty=repetition_penalty,
+            no_repeat_ngram_size=no_repeat_ngram_size,
             log_prob_threshold=log_prob_threshold,
             no_speech_threshold=no_speech_threshold,
             compression_ratio_threshold=compression_ratio_threshold,
             condition_on_previous_text=condition_on_previous_text,
+            prompt_reset_on_temperature=prompt_reset_on_temperature,
             temperatures=(
                 temperature if isinstance(temperature, (list, tuple)) else [temperature]
             ),
@@ -331,6 +438,9 @@ class WhisperModel:
             word_timestamps=word_timestamps,
             prepend_punctuations=prepend_punctuations,
             append_punctuations=append_punctuations,
+            max_new_tokens=max_new_tokens,
+            clip_timestamps=clip_timestamps,
+            hallucination_silence_threshold=hallucination_silence_threshold,
         )
 
         segments = self.generate_segments(features, tokenizer, options, encoder_output)
@@ -342,6 +452,7 @@ class WhisperModel:
             language=language,
             language_probability=language_probability,
             duration=duration,
+            duration_after_vad=duration_after_vad,
             transcription_options=options,
             vad_options=vad_parameters,
             all_language_probs=all_language_probs,
@@ -357,8 +468,33 @@ class WhisperModel:
         encoder_output: Optional[ctranslate2.StorageView] = None,
     ) -> Iterable[Segment]:
         content_frames = features.shape[-1] - self.feature_extractor.nb_max_frames
+        content_duration = float(content_frames * self.feature_extractor.time_per_frame)
+
+        if isinstance(options.clip_timestamps, str):
+            TranscriptionOptions.clip_timestamps = [
+                float(ts)
+                for ts in (
+                    options.clip_timestamps.split(",")
+                    if options.clip_timestamps
+                    else []
+                )
+            ]
+        seek_points: List[int] = [
+            round(ts * self.frames_per_second) for ts in options.clip_timestamps
+        ]
+        if len(seek_points) == 0:
+            seek_points.append(0)
+        if len(seek_points) % 2 == 1:
+            seek_points.append(content_frames)
+        seek_clips: List[Tuple[int, int]] = list(
+            zip(seek_points[::2], seek_points[1::2])
+        )
+
+        punctuation = "\"'“¿([{-\"'.。,，!！?？:：”)]}、"
+
         idx = 0
-        seek = 0
+        clip_idx = 0
+        seek = seek_clips[clip_idx][0]
         all_tokens = []
         prompt_reset_since = 0
 
@@ -371,13 +507,34 @@ class WhisperModel:
                 all_tokens.extend(options.initial_prompt)
 
         last_speech_timestamp = 0.0
-        while seek < content_frames:
+        # NOTE: This loop is obscurely flattened to make the diff readable.
+        # A later commit should turn this into a simpler nested loop.
+        # for seek_clip_start, seek_clip_end in seek_clips:
+        #     while seek < seek_clip_end
+        while clip_idx < len(seek_clips):
+            seek_clip_start, seek_clip_end = seek_clips[clip_idx]
+            if seek_clip_end > content_frames:
+                seek_clip_end = content_frames
+            if seek < seek_clip_start:
+                seek = seek_clip_start
+            if seek >= seek_clip_end:
+                clip_idx += 1
+                if clip_idx < len(seek_clips):
+                    seek = seek_clips[clip_idx][0]
+                continue
             time_offset = seek * self.feature_extractor.time_per_frame
-            segment = features[:, seek : seek + self.feature_extractor.nb_max_frames]
-            segment_size = min(
-                self.feature_extractor.nb_max_frames, content_frames - seek
+            window_end_time = float(
+                (seek + self.feature_extractor.nb_max_frames)
+                * self.feature_extractor.time_per_frame
             )
+            segment_size = min(
+                self.feature_extractor.nb_max_frames,
+                content_frames - seek,
+                seek_clip_end - seek,
+            )
+            segment = features[:, seek : seek + segment_size]
             segment_duration = segment_size * self.feature_extractor.time_per_frame
+            segment = pad_or_trim(segment, self.feature_extractor.nb_max_frames)
 
             if self.logger.isEnabledFor(logging.DEBUG):
                 self.logger.debug(
@@ -392,7 +549,7 @@ class WhisperModel:
                 prefix=options.prefix if seek == 0 else None,
             )
 
-            if encoder_output is None:
+            if seek > 0 or encoder_output is None:
                 encoder_output = self.encode(segment)
 
             (
@@ -429,10 +586,33 @@ class WhisperModel:
             previous_seek = seek
             current_segments = []
 
+            # anomalous words are very long/short/improbable
+            def word_anomaly_score(word: dict) -> float:
+                probability = word.get("probability", 0.0)
+                duration = word["end"] - word["start"]
+                score = 0.0
+                if probability < 0.15:
+                    score += 1.0
+                if duration < 0.133:
+                    score += (0.133 - duration) * 15
+                if duration > 2.0:
+                    score += duration - 2.0
+                return score
+
+            def is_segment_anomaly(segment: Optional[dict]) -> bool:
+                if segment is None or not segment["words"]:
+                    return False
+                words = [w for w in segment["words"] if w["word"] not in punctuation]
+                words = words[:8]
+                score = sum(word_anomaly_score(w) for w in words)
+                return score >= 3 or score + 0.01 >= len(words)
+
+            def next_words_segment(segments: List[dict]) -> Optional[dict]:
+                return next((s for s in segments if s["words"]), None)
+
             single_timestamp_ending = (
                 len(tokens) >= 2
-                and tokens[-2] < tokenizer.timestamp_begin
-                and tokens[-1] >= tokenizer.timestamp_begin
+                and tokens[-2] < tokenizer.timestamp_begin <= tokens[-1]
             )
 
             consecutive_timestamps = [
@@ -515,20 +695,62 @@ class WhisperModel:
                     last_speech_timestamp=last_speech_timestamp,
                 )
 
-                word_end_timestamps = [
-                    w["end"] for s in current_segments for w in s["words"]
-                ]
-                if len(word_end_timestamps) > 0:
-                    last_speech_timestamp = word_end_timestamps[-1]
-                if not single_timestamp_ending and len(word_end_timestamps) > 0:
-                    seek_shift = round(
-                        (word_end_timestamps[-1] - time_offset) * self.frames_per_second
-                    )
+                if not single_timestamp_ending:
+                    last_word_end = get_end(current_segments)
+                    if last_word_end is not None and last_word_end > time_offset:
+                        seek = round(last_word_end * self.frames_per_second)
 
-                    if seek_shift > 0:
-                        seek = previous_seek + seek_shift
+                # skip silence before possible hallucinations
+                if options.hallucination_silence_threshold is not None:
+                    threshold = options.hallucination_silence_threshold
 
-            encoder_output = None
+                    # if first segment might be a hallucination, skip leading silence
+                    first_segment = next_words_segment(current_segments)
+                    if first_segment is not None and is_segment_anomaly(first_segment):
+                        gap = first_segment["start"] - time_offset
+                        if gap > threshold:
+                            seek = previous_seek + round(gap * self.frames_per_second)
+                            continue
+
+                    # skip silence before any possible hallucination that is surrounded
+                    # by silence or more hallucinations
+                    hal_last_end = last_speech_timestamp
+                    for si in range(len(current_segments)):
+                        segment = current_segments[si]
+                        if not segment["words"]:
+                            continue
+                        if is_segment_anomaly(segment):
+                            next_segment = next_words_segment(
+                                current_segments[si + 1 :]
+                            )
+                            if next_segment is not None:
+                                hal_next_start = next_segment["words"][0]["start"]
+                            else:
+                                hal_next_start = time_offset + segment_duration
+                            silence_before = (
+                                segment["start"] - hal_last_end > threshold
+                                or segment["start"] < threshold
+                                or segment["start"] - time_offset < 2.0
+                            )
+                            silence_after = (
+                                hal_next_start - segment["end"] > threshold
+                                or is_segment_anomaly(next_segment)
+                                or window_end_time - segment["end"] < 2.0
+                            )
+                            if silence_before and silence_after:
+                                seek = round(
+                                    max(time_offset + 1, segment["start"])
+                                    * self.frames_per_second
+                                )
+                                if content_duration - segment["end"] < threshold:
+                                    seek = content_frames
+                                current_segments[si:] = []
+                                break
+                        hal_last_end = segment["end"]
+
+                last_word_end = get_end(current_segments)
+                if last_word_end is not None:
+                    last_speech_timestamp = last_word_end
 
             for segment in current_segments:
                 tokens = segment["tokens"]
@@ -558,7 +780,17 @@ class WhisperModel:
                     ),
                 )
 
-            if not options.condition_on_previous_text or temperature > 0.5:
+            if (
+                not options.condition_on_previous_text
+                or temperature > options.prompt_reset_on_temperature
+            ):
+                if options.condition_on_previous_text:
+                    self.logger.debug(
+                        "Reset prompt. prompt_reset_on_temperature threshold is met %f > %f",
+                        temperature,
+                        options.prompt_reset_on_temperature,
+                    )
+
                 prompt_reset_since = len(all_tokens)
 
     def encode(self, features: np.ndarray) -> ctranslate2.StorageView:
@@ -578,14 +810,28 @@ class WhisperModel:
         tokenizer: Tokenizer,
         options: TranscriptionOptions,
     ) -> Tuple[ctranslate2.models.WhisperGenerationResult, float, float, float]:
-        result = None
-        avg_logprob = None
-        final_temperature = None
-        compression_ratio = None
+        decode_result = None
+        all_results = []
+        below_cr_threshold_results = []
 
         max_initial_timestamp_index = int(
             round(options.max_initial_timestamp / self.time_precision)
         )
+        if options.max_new_tokens is not None:
+            max_length = len(prompt) + options.max_new_tokens
+        else:
+            max_length = self.max_length
+
+        if max_length > self.max_length:
+            raise ValueError(
+                f"The length of the prompt is {len(prompt)}, and the `max_new_tokens` "
+                f"{max_length - len(prompt)}. Thus, the combined length of the prompt "
+                f"and `max_new_tokens` is: {max_length}. This exceeds the "
+                f"`max_length` of the Whisper model: {self.max_length}. "
+                "You should either reduce the length of your prompt, or "
+                "reduce the value of `max_new_tokens`, "
+                f"so that their combined length is less that {self.max_length}."
+            )
 
         for temperature in options.temperatures:
             if temperature > 0:
@@ -601,12 +847,13 @@ class WhisperModel:
                     "patience": options.patience,
                 }
 
-            final_temperature = temperature
             result = self.model.generate(
                 encoder_output,
                 [prompt],
                 length_penalty=options.length_penalty,
-                max_length=self.max_length,
+                repetition_penalty=options.repetition_penalty,
+                no_repeat_ngram_size=options.no_repeat_ngram_size,
+                max_length=max_length,
                 return_scores=True,
                 return_no_speech_prob=True,
                 suppress_blank=options.suppress_blank,
@@ -625,20 +872,28 @@ class WhisperModel:
             text = tokenizer.decode(tokens).strip()
             compression_ratio = get_compression_ratio(text)
 
+            decode_result = (
+                result,
+                avg_logprob,
+                temperature,
+                compression_ratio,
+            )
+            all_results.append(decode_result)
+
             needs_fallback = False
 
-            if (
-                options.compression_ratio_threshold is not None
-                and compression_ratio > options.compression_ratio_threshold
-            ):
-                needs_fallback = True  # too repetitive
+            if options.compression_ratio_threshold is not None:
+                if compression_ratio > options.compression_ratio_threshold:
+                    needs_fallback = True  # too repetitive
 
-                self.logger.debug(
-                    "Compression ratio threshold is not met with temperature %.1f (%f > %f)",
-                    temperature,
-                    compression_ratio,
-                    options.compression_ratio_threshold,
-                )
+                    self.logger.debug(
+                        "Compression ratio threshold is not met with temperature %.1f (%f > %f)",
+                        temperature,
+                        compression_ratio,
+                        options.compression_ratio_threshold,
+                    )
+                else:
+                    below_cr_threshold_results.append(decode_result)
 
             if (
                 options.log_prob_threshold is not None
@@ -656,13 +911,27 @@ class WhisperModel:
             if (
                 options.no_speech_threshold is not None
                 and result.no_speech_prob > options.no_speech_threshold
+                and options.log_prob_threshold is not None
+                and avg_logprob < options.log_prob_threshold
             ):
                 needs_fallback = False  # silence
 
             if not needs_fallback:
                 break
+        else:
+            # all failed, select the result with the highest average log probability
+            decode_result = max(
+                below_cr_threshold_results or all_results, key=lambda x: x[1]
+            )
+            # to pass final temperature for prompt_reset_on_temperature
+            decode_result = (
+                decode_result[0],
+                decode_result[1],
+                temperature,
+                decode_result[3],
+            )
 
-        return result, avg_logprob, final_temperature, compression_ratio
+        return decode_result
 
     def get_prompt(
         self,
@@ -686,6 +955,8 @@ class WhisperModel:
             prefix_tokens = tokenizer.encode(" " + prefix.strip())
             if len(prefix_tokens) >= self.max_length // 2:
                 prefix_tokens = prefix_tokens[: self.max_length // 2 - 1]
+            if not without_timestamps:
+                prompt.append(tokenizer.timestamp_begin)
             prompt.extend(prefix_tokens)
 
         return prompt
@@ -699,7 +970,7 @@ class WhisperModel:
         prepend_punctuations: str,
         append_punctuations: str,
         last_speech_timestamp: float,
-    ):
+    ) -> None:
         if len(segments) == 0:
             return
 
@@ -715,13 +986,12 @@ class WhisperModel:
         word_durations = np.array([word["end"] - word["start"] for word in alignment])
         word_durations = word_durations[word_durations.nonzero()]
         median_duration = np.median(word_durations) if len(word_durations) > 0 else 0.0
+        median_duration = min(0.7, float(median_duration))
         max_duration = median_duration * 2
 
         # hack: truncate long words at sentence boundaries.
         # a better segmentation algorithm based on VAD should be able to replace this.
         if len(word_durations) > 0:
-            median_duration = np.median(word_durations)
-            max_duration = median_duration * 2
             sentence_end_marks = ".。!！?？"
             # ensure words at sentence boundaries
             # are not longer than twice the median word duration.
@@ -838,6 +1108,13 @@ class WhisperModel:
         words, word_tokens = tokenizer.split_to_word_tokens(
             text_tokens + [tokenizer.eot]
         )
+        if len(word_tokens) <= 1:
+            # return on eot only
+            # >>> np.pad([], (1, 0))
+            # array([0.])
+            # This results in crashes when we lookup jump_times with float, like
+            # IndexError: arrays used as indices must be of integer (or boolean) type
+            return []
         word_boundaries = np.pad(np.cumsum([len(t) for t in word_tokens[:-1]]), (1, 0))
         if len(word_boundaries) <= 1:
             return []
@@ -907,7 +1184,10 @@ def get_compression_ratio(text: str) -> float:
     return len(text_bytes) / len(zlib.compress(text_bytes))
 
 
-def get_suppressed_tokens(tokenizer, suppress_tokens):
+def get_suppressed_tokens(
+    tokenizer: Tokenizer,
+    suppress_tokens: Optional[List[int]],
+) -> Optional[List[int]]:
     if not suppress_tokens or -1 in suppress_tokens:
         return suppress_tokens
 
@@ -928,7 +1208,7 @@ def get_suppressed_tokens(tokenizer, suppress_tokens):
     return sorted(set(suppress_tokens))
 
 
-def merge_punctuations(alignment: List[dict], prepended: str, appended: str):
+def merge_punctuations(alignment: List[dict], prepended: str, appended: str) -> None:
     # merge prepended punctuations
     i = len(alignment) - 2
     j = len(alignment) - 1

faster_whisper/utils.py

@@ -2,25 +2,36 @@ import logging
 import os
 import re
 
-from typing import Optional
+from typing import List, Optional
 
 import huggingface_hub
 import requests
 
 from tqdm.auto import tqdm
 
-_MODELS = (
-    "tiny.en",
-    "tiny",
-    "base.en",
-    "base",
-    "small.en",
-    "small",
-    "medium.en",
-    "medium",
-    "large-v1",
-    "large-v2",
-)
+_MODELS = {
+    "tiny.en": "Systran/faster-whisper-tiny.en",
+    "tiny": "Systran/faster-whisper-tiny",
+    "base.en": "Systran/faster-whisper-base.en",
+    "base": "Systran/faster-whisper-base",
+    "small.en": "Systran/faster-whisper-small.en",
+    "small": "Systran/faster-whisper-small",
+    "medium.en": "Systran/faster-whisper-medium.en",
+    "medium": "Systran/faster-whisper-medium",
+    "large-v1": "Systran/faster-whisper-large-v1",
+    "large-v2": "Systran/faster-whisper-large-v2",
+    "large-v3": "Systran/faster-whisper-large-v3",
+    "large": "Systran/faster-whisper-large-v3",
+    "distil-large-v2": "Systran/faster-distil-whisper-large-v2",
+    "distil-medium.en": "Systran/faster-distil-whisper-medium.en",
+    "distil-small.en": "Systran/faster-distil-whisper-small.en",
+    "distil-large-v3": "Systran/faster-distil-whisper-large-v3",
+}
+
+
+def available_models() -> List[str]:
+    """Returns the names of available models."""
+    return list(_MODELS.keys())
 
 
 def get_assets_path():
@@ -41,12 +52,11 @@ def download_model(
 ):
     """Downloads a CTranslate2 Whisper model from the Hugging Face Hub.
 
-    The model is downloaded from https://huggingface.co/guillaumekln.
-
     Args:
-      size_or_id: Size of the model to download (tiny, tiny.en, base, base.en, small, small.en,
-        medium, medium.en, large-v1, or large-v2), or a CTranslate2-converted model ID
-        from the Hugging Face Hub (e.g. guillaumekln/faster-whisper-large-v2).
+      size_or_id: Size of the model to download from https://huggingface.co/Systran
+        (tiny, tiny.en, base, base.en, small, small.en medium, medium.en, large-v1, large-v2,
+        large-v3, large), or a CTranslate2-converted model ID from the Hugging Face Hub
+        (e.g. Systran/faster-whisper-large-v3).
       output_dir: Directory where the model should be saved. If not set, the model is saved in
         the cache directory.
       local_files_only:  If True, avoid downloading the file and return the path to the local
@@ -62,16 +72,16 @@ def download_model(
     if re.match(r".*/.*", size_or_id):
         repo_id = size_or_id
     else:
-        if size_or_id not in _MODELS:
+        repo_id = _MODELS.get(size_or_id)
+        if repo_id is None:
             raise ValueError(
                 "Invalid model size '%s', expected one of: %s"
-                % (size_or_id, ", ".join(_MODELS))
+                % (size_or_id, ", ".join(_MODELS.keys()))
             )
 
-        repo_id = "guillaumekln/faster-whisper-%s" % size_or_id
-
     allow_patterns = [
         "config.json",
+        "preprocessor_config.json",
         "model.bin",
         "tokenizer.json",
         "vocabulary.*",
@@ -137,3 +147,10 @@ class disabled_tqdm(tqdm):
     def __init__(self, *args, **kwargs):
         kwargs["disable"] = True
         super().__init__(*args, **kwargs)
+
+
+def get_end(segments: List[dict]) -> Optional[float]:
+    return next(
+        (w["end"] for s in reversed(segments) for w in reversed(s["words"])),
+        segments[-1]["end"] if segments else None,
+    )

faster_whisper/version.py

@@ -1,3 +1,3 @@
 """Version information."""
 
-__version__ = "0.6.0"
+__version__ = "1.0.1"

requirements.txt

@@ -1,5 +1,5 @@
-av==10.*
-ctranslate2>=3.10,<4
+av==11.*
+ctranslate2>=4.0,<5
 huggingface_hub>=0.13
-tokenizers==0.13.*
+tokenizers>=0.13,<0.16
 onnxruntime>=1.14,<2

setup.py

@@ -37,7 +37,7 @@ setup(
     long_description=get_long_description(),
     long_description_content_type="text/markdown",
     author="Guillaume Klein",
-    url="https://github.com/guillaumekln/faster-whisper",
+    url="https://github.com/SYSTRAN/faster-whisper",
     classifiers=[
         "Development Status :: 4 - Beta",
         "Intended Audience :: Developers",

tests/test_transcribe.py

@@ -3,6 +3,11 @@ import os
 from faster_whisper import WhisperModel, decode_audio
 
 
+def test_supported_languages():
+    model = WhisperModel("tiny.en")
+    assert model.supported_languages == ["en"]
+
+
 def test_transcribe(jfk_path):
     model = WhisperModel("tiny")
     segments, info = model.transcribe(jfk_path, word_timestamps=True)
@@ -34,6 +39,24 @@ def test_transcribe(jfk_path):
     assert segment.end == segment.words[-1].end
 
 
+def test_prefix_with_timestamps(jfk_path):
+    model = WhisperModel("tiny")
+    segments, _ = model.transcribe(jfk_path, prefix="And so my fellow Americans")
+    segments = list(segments)
+
+    assert len(segments) == 1
+
+    segment = segments[0]
+
+    assert segment.text == (
+        " And so my fellow Americans ask not what your country can do for you, "
+        "ask what you can do for your country."
+    )
+
+    assert segment.start == 0
+    assert 10 < segment.end < 11
+
+
 def test_vad(jfk_path):
     model = WhisperModel("tiny")
     segments, info = model.transcribe(

tests/test_utils.py

@@ -1,6 +1,12 @@
 import os
 
-from faster_whisper import download_model
+from faster_whisper import available_models, download_model
+
+
+def test_available_models():
+    models = available_models()
+    assert isinstance(models, list)
+    assert "tiny" in models
 
 
 def test_download_model(tmpdir):