diff --git a/MANIFEST.in b/MANIFEST.in index 8a103dd..b33ab28 100644 --- a/MANIFEST.in +++ b/MANIFEST.in @@ -1,4 +1,4 @@ -include faster_whisper/assets/silero_vad.onnx +include faster_whisper/assets/silero_encoder_v5.onnx +include faster_whisper/assets/silero_decoder_v5.onnx include requirements.txt include requirements.conversion.txt -include faster_whisper/assets/pyannote_vad_model.bin diff --git a/README.md b/README.md index f7d54ee..7cb9605 100644 --- a/README.md +++ b/README.md @@ -178,9 +178,6 @@ language_info = model.detect_language_multi_segment("audio.mp3") ### Batched faster-whisper - -The batched version of faster-whisper is inspired by [whisper-x](https://github.com/m-bain/whisperX) licensed under the BSD-2 Clause license and integrates its VAD model to this library. We modify this implementation and also replaced the feature extraction with a faster torch-based implementation. Batched version improves the speed upto 10-12x compared to openAI implementation and 3-4x compared to the sequential faster_whisper version. It works by transcribing semantically meaningful audio chunks as batches leading to faster inference. - The following code snippet illustrates how to run inference with batched version on an example audio file. Please also refer to the test scripts of batched faster whisper. ```python diff --git a/benchmark/evaluate_yt_commons.py b/benchmark/evaluate_yt_commons.py new file mode 100644 index 0000000..0511be6 --- /dev/null +++ b/benchmark/evaluate_yt_commons.py @@ -0,0 +1,83 @@ +import argparse +import json +import os + +from io import BytesIO + +from datasets import load_dataset +from evaluate import load +from pytubefix import YouTube +from torch.utils.data import DataLoader +from tqdm import tqdm +from transformers.models.whisper.english_normalizer import EnglishTextNormalizer + +from faster_whisper import BatchedInferencePipeline, WhisperModel, decode_audio + + +def url_to_audio(row): + buffer = BytesIO() + yt = YouTube(row["link"]) + video = ( + yt.streams.filter(only_audio=True, mime_type="audio/mp4") + .order_by("bitrate") + .desc() + .first() + ) + video.stream_to_buffer(buffer) + buffer.seek(0) + row["audio"] = decode_audio(buffer) + return row + + +parser = argparse.ArgumentParser(description="WER benchmark") +parser.add_argument( + "--audio_numb", + type=int, + default=None, + help="Specify the number of validation audio files in the dataset." + " Set to None to retrieve all audio files.", +) +args = parser.parse_args() + +# define the evaluation metric +wer_metric = load("wer") + +with open(os.path.join(os.path.dirname(__file__), "normalizer.json"), "r") as f: + normalizer = EnglishTextNormalizer(json.load(f)) + +dataset = load_dataset("mobiuslabsgmbh/youtube-commons-asr-eval", streaming=True).map( + url_to_audio +) +dataset = iter( + DataLoader(dataset["test"], batch_size=1, prefetch_factor=4, num_workers=2) +) + +model = WhisperModel("large-v3", device="cuda") +pipeline = BatchedInferencePipeline(model, device="cuda") + + +all_transcriptions = [] +all_references = [] +# iterate over the dataset and run inference +for i, row in tqdm(enumerate(dataset), desc="Evaluating..."): + result, info = pipeline.transcribe( + row["audio"][0], + batch_size=8, + word_timestamps=False, + without_timestamps=True, + ) + + all_transcriptions.append("".join(segment.text for segment in result)) + all_references.append(row["text"][0]) + if args.audio_numb and i == (args.audio_numb - 1): + break + +# normalize predictions and references +all_transcriptions = [normalizer(transcription) for transcription in all_transcriptions] +all_references = [normalizer(reference) for reference in all_references] + +# compute the WER metric +wer = 100 * wer_metric.compute( + predictions=all_transcriptions, references=all_references +) +print("WER: %.3f" % wer) diff --git a/benchmark/requirements.benchmark.txt b/benchmark/requirements.benchmark.txt index 2645004..c49dcca 100644 --- a/benchmark/requirements.benchmark.txt +++ b/benchmark/requirements.benchmark.txt @@ -4,3 +4,4 @@ evaluate datasets memory_profiler py3nvml +pytubefix diff --git a/faster_whisper/assets/pyannote_vad_model.bin b/faster_whisper/assets/pyannote_vad_model.bin deleted file mode 100644 index 75c92f0..0000000 Binary files a/faster_whisper/assets/pyannote_vad_model.bin and /dev/null differ diff --git a/faster_whisper/assets/silero_decoder_v5.onnx b/faster_whisper/assets/silero_decoder_v5.onnx new file mode 100644 index 0000000..0097e74 Binary files /dev/null and b/faster_whisper/assets/silero_decoder_v5.onnx differ diff --git a/faster_whisper/assets/silero_encoder_v5.onnx b/faster_whisper/assets/silero_encoder_v5.onnx new file mode 100644 index 0000000..0c073e2 Binary files /dev/null and b/faster_whisper/assets/silero_encoder_v5.onnx differ diff --git a/faster_whisper/assets/silero_vad.onnx b/faster_whisper/assets/silero_vad.onnx deleted file mode 100644 index d0ccd9d..0000000 Binary files a/faster_whisper/assets/silero_vad.onnx and /dev/null differ diff --git a/faster_whisper/transcribe.py b/faster_whisper/transcribe.py index 8652e82..7611237 100644 --- a/faster_whisper/transcribe.py +++ b/faster_whisper/transcribe.py @@ -7,6 +7,7 @@ import zlib from collections import Counter, defaultdict from inspect import signature +from math import ceil from typing import BinaryIO, Iterable, List, NamedTuple, Optional, Tuple, Union import ctranslate2 @@ -14,26 +15,18 @@ import numpy as np import tokenizers import torch -from pyannote.audio import Model from tqdm import tqdm from faster_whisper.audio import decode_audio, pad_or_trim from faster_whisper.feature_extractor import FeatureExtractor from faster_whisper.tokenizer import _LANGUAGE_CODES, Tokenizer -from faster_whisper.utils import ( - download_model, - format_timestamp, - get_assets_path, - get_end, - get_logger, -) +from faster_whisper.utils import download_model, format_timestamp, get_end, get_logger from faster_whisper.vad import ( SpeechTimestampsMap, VadOptions, - VoiceActivitySegmentation, collect_chunks, get_speech_timestamps, - merge_chunks, + merge_segments, ) @@ -115,67 +108,26 @@ class BatchedInferencePipeline: def __init__( self, model, - use_vad_model: bool = True, options: Optional[NamedTuple] = None, tokenizer=None, - chunk_length: int = 30, - vad_device: Union[int, str, "torch.device"] = "auto", - vad_onset: float = 0.500, - vad_offset: float = 0.363, language: Optional[str] = None, ): self.model: WhisperModel = model self.tokenizer = tokenizer self.options = options self.preset_language = language - self.use_vad_model = use_vad_model - self.vad_onset = vad_onset - self.vad_offset = vad_offset - self.vad_model_path = os.path.join(get_assets_path(), "pyannote_vad_model.bin") - if self.use_vad_model: - self.vad_device = self.get_device(vad_device) - self.vad_model = self.load_vad_model( - vad_onset=self.vad_onset, vad_offset=self.vad_offset - ) - else: - self.vad_model = None - self.chunk_length = chunk_length # VAD merging size self.last_speech_timestamp = 0.0 - def get_device(self, device: Union[int, str, "torch.device"]): - """ - Converts the input device into a torch.device object. - - The input can be an integer, a string, or a `torch.device` object. - - The function handles a special case where the input device is "auto". - When "auto" is specified, the device will default to the - device of the model (self.model.device). If the model's device is also "auto", - it selects "cuda" if a CUDA-capable device is available; otherwise, it selects "cpu". - """ - if isinstance(device, torch.device): - return device - elif isinstance(device, str): - if device == "auto" and self.model.device == "auto": - device = "cuda" if torch.cuda.is_available() else "cpu" - elif device == "auto": - device = self.model.device - return torch.device(device) - elif device < 0: - return torch.device("cpu") - else: - return torch.device(f"cuda:{device}") - - def forward(self, features, segments_metadata, **forward_params): + def forward(self, features, chunks_metadata, **forward_params): encoder_output, outputs = self.model.generate_segment_batched( features, self.tokenizer, forward_params ) segmented_outputs = [] segment_sizes = [] - for segment_metadata, output in zip(segments_metadata, outputs): - duration = segment_metadata["end_time"] - segment_metadata["start_time"] - segment_size = int(duration * self.model.frames_per_second) + for chunk_metadata, output in zip(chunks_metadata, outputs): + duration = chunk_metadata["end_time"] - chunk_metadata["start_time"] + segment_size = int(ceil(duration) * self.model.frames_per_second) segment_sizes.append(segment_size) ( subsegments, @@ -184,7 +136,7 @@ class BatchedInferencePipeline: ) = self.model._split_segments_by_timestamps( tokenizer=self.tokenizer, tokens=output["tokens"], - time_offset=segment_metadata["start_time"], + time_offset=chunk_metadata["start_time"], segment_size=segment_size, segment_duration=duration, seek=0, @@ -252,43 +204,9 @@ class BatchedInferencePipeline: return language, language_probability, task, all_language_probs - @staticmethod - def audio_split(audio, segments, sampling_rate): - """Returns splitted audio chunks as iterator""" - audio_segments = [] - segments_metadata = [] - for seg in segments: - f1 = int(seg["start"] * sampling_rate) - f2 = int(seg["end"] * sampling_rate) - seg_metadata = { - "start_time": seg["start"], - "end_time": seg["end"], - "stitched_seg": seg["segments"], - } - audio_segments.append(audio[f1:f2]) - segments_metadata.append(seg_metadata) - return audio_segments, segments_metadata - - def load_vad_model(self, vad_onset=0.500, vad_offset=0.363): - vad_model = Model.from_pretrained(self.vad_model_path) - hyperparameters = { - "onset": vad_onset, - "offset": vad_offset, - "min_duration_on": 0.1, - "min_duration_off": 0.1, - } - - vad_pipeline = VoiceActivitySegmentation( - segmentation=vad_model, device=torch.device(self.vad_device) - ) - vad_pipeline.instantiate(hyperparameters) - return vad_pipeline - def transcribe( self, - audio: Union[str, torch.Tensor, np.ndarray], - vad_segments: Optional[List[dict]] = None, - batch_size: int = 16, + audio: Union[str, BinaryIO, torch.Tensor, np.ndarray], language: Optional[str] = None, task: str = None, log_progress: bool = False, @@ -314,26 +232,26 @@ class BatchedInferencePipeline: prefix: Optional[str] = None, suppress_blank: bool = True, suppress_tokens: Optional[List[int]] = [-1], + without_timestamps: bool = True, + word_timestamps: bool = False, prepend_punctuations: str = "\"'“¿([{-", append_punctuations: str = "\"'.。,，!！?？:：”)]}、", + vad_filter: bool = True, + vad_parameters: Optional[Union[dict, VadOptions]] = None, max_new_tokens: Optional[int] = None, + chunk_length: Optional[int] = None, + clip_timestamps: Optional[List[dict]] = None, + batch_size: int = 16, hotwords: Optional[str] = None, - word_timestamps: bool = False, - without_timestamps: bool = True, ) -> Tuple[Iterable[Segment], TranscriptionInfo]: """transcribe audio in chunks in batched fashion and return with language info. Arguments: - audio: audio file as numpy array/path for batched transcription. - vad_segments: Optionally provide list of dictionaries each containing "start", "end", - and "segments" keys. - "start" and "end" keys specify the start and end of the voiced region within - 30 sec boundary. An additional key "segments" contains all the start - and end of voiced regions within that 30sec boundary as a list of tuples. - If no vad_segments specified, it uses internal vad model automatically segment them. - batch_size: the maximum number of parallel requests to model for decoding. - language: The language spoken in the audio. - task: either "transcribe" or "translate". + audio: Path to the input file (or a file-like object), or the audio waveform. + language: The language spoken in the audio. It should be a language code such + as "en" or "fr". If not set, the language will be detected in the first 30 seconds + of audio. + task: Task to execute (transcribe or translate). log_progress: whether to show progress bar or not. beam_size: Beam size to use for decoding. best_of: Number of candidates when sampling with non-zero temperature. @@ -350,8 +268,8 @@ class BatchedInferencePipeline: log_prob_threshold: If the average log probability over sampled tokens is below this value, treat as failed. log_prob_low_threshold: This parameter alone is sufficient to skip an output text, - whereas log_prob_threshold also looks for appropriate no_speech_threshold value. - This value should be less than log_prob_threshold. + whereas log_prob_threshold also looks for appropriate no_speech_threshold value. + This value should be less than log_prob_threshold. no_speech_threshold: If the no_speech probability is higher than this value AND the average log probability over sampled tokens is below `log_prob_threshold`, consider the segment as silent. @@ -361,18 +279,29 @@ class BatchedInferencePipeline: suppress_blank: Suppress blank outputs at the beginning of the sampling. suppress_tokens: List of token IDs to suppress. -1 will suppress a default set of symbols as defined in `tokenizer.non_speech_tokens()`. + without_timestamps: Only sample text tokens. + word_timestamps: Extract word-level timestamps using the cross-attention pattern + and dynamic time warping, and include the timestamps for each word in each segment. + Set as False. prepend_punctuations: If word_timestamps is True, merge these punctuation symbols with the next word append_punctuations: If word_timestamps is True, merge these punctuation symbols with the previous word + vad_filter: Enable the voice activity detection (VAD) to filter out parts of the audio + without speech. This step is using the Silero VAD model + 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: Optionally provide list of dictionaries each containing "start" and + "end" keys that specify the start and end of the voiced region within + `chunk_length` boundary. vad_filter will be ignored if clip_timestamps is used. + batch_size: the maximum number of parallel requests to model for decoding. hotwords: Hotwords/hint phrases to the model. Has no effect if prefix is not None. - word_timestamps: Extract word-level timestamps using the cross-attention pattern - and dynamic time warping, and include the timestamps for each word in each segment. - Set as False. - without_timestamps: Only sample text tokens. Static params: (Fixed for batched version) max_initial_timestamp: The initial timestamp cannot be later than this, set at 0.0. @@ -390,28 +319,18 @@ class BatchedInferencePipeline: 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. set as None. - clip_timestamps: - 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. Set as "0". unused: 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. - vad_filter: Enable the voice activity detection (VAD) to filter out parts of the audio - without speech. This step is using the Silero VAD model - 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`). - chunk_length: The length of audio segments. If it is not None, it will overwrite the - default chunk_length of the FeatureExtractor. Returns: A tuple with: - - a generator over transcribed batched segments. - - an instance of TranscriptionInfo. + - a generator over transcribed segments + - an instance of TranscriptionInfo """ sampling_rate = self.model.feature_extractor.sampling_rate @@ -422,29 +341,32 @@ class BatchedInferencePipeline: audio = decode_audio(audio, sampling_rate=sampling_rate) duration = audio.shape[0] / sampling_rate + chunk_length = chunk_length or self.model.feature_extractor.chunk_length # if no segment split is provided, use vad_model and generate segments - if not vad_segments: - # run the audio if it is less than 30 sec even without vad_segments - if self.use_vad_model: - vad_segments = self.vad_model( - { - "waveform": audio.unsqueeze(0), - "sample_rate": 16000, - } - ) - vad_segments = merge_chunks( - vad_segments, - self.chunk_length, - onset=self.vad_onset, - offset=self.vad_offset, - ) - elif duration < self.chunk_length: - vad_segments = [ - {"start": 0.0, "end": duration, "segments": [(0.0, duration)]} - ] + if not clip_timestamps: + if vad_filter: + if vad_parameters is None: + vad_parameters = VadOptions( + max_speech_duration_s=chunk_length, + min_silence_duration_ms=160, + ) + elif isinstance(vad_parameters, dict): + if "max_speech_duration_s" in vad_parameters.keys(): + vad_parameters.pop("max_speech_duration_s") + + vad_parameters = VadOptions( + **vad_parameters, max_speech_duration_s=chunk_length + ) + + active_segments = get_speech_timestamps(audio, vad_parameters) + clip_timestamps = merge_segments(active_segments, vad_parameters) + # run the audio if it is less than 30 sec even without clip_timestamps + elif duration < chunk_length: + clip_timestamps = [{"start": 0, "end": audio.shape[0]}] else: raise RuntimeError( - "No vad segments found. Set 'use_vad_model' to True while loading the model" + "No clip timestamps found. " + "Set 'vad_filter' to True or provide 'clip_timestamps'." ) if self.model.model.is_multilingual: language = language or self.preset_language @@ -452,7 +374,7 @@ class BatchedInferencePipeline: if language is not None: self.model.logger.warning( f"English-only model is used, but {language} language is" - "chosen, setting language to 'en'." + " chosen, setting language to 'en'." ) language = "en" @@ -463,8 +385,9 @@ class BatchedInferencePipeline: all_language_probs, ) = self.get_language_and_tokenizer(audio, task, language) - duration_after_vad = sum( - segment["end"] - segment["start"] for segment in vad_segments + duration_after_vad = ( + sum((segment["end"] - segment["start"]) for segment in clip_timestamps) + / sampling_rate ) # batched options: see the difference with default options in WhisperModel @@ -511,27 +434,26 @@ class BatchedInferencePipeline: all_language_probs=all_language_probs, ) - audio_segments, segments_metadata = self.audio_split( - audio, vad_segments, sampling_rate - ) + audio_chunks, chunks_metadata = collect_chunks(audio, clip_timestamps) to_cpu = ( self.model.model.device == "cuda" and len(self.model.model.device_index) > 1 ) - audio_segments = torch.nested.nested_tensor(audio_segments).to_padded_tensor( - padding=0 - ) - features = torch.stack( - [ - self.model.feature_extractor(audio_segment, to_cpu=to_cpu)[ - ..., : self.model.feature_extractor.nb_max_frames + features = ( + torch.stack( + [ + self.model.feature_extractor(chunk, to_cpu=to_cpu)[ + ..., : self.model.feature_extractor.nb_max_frames + ] + for chunk in audio_chunks ] - for audio_segment in audio_segments - ] + ) + if duration_after_vad + else [] ) segments = self._batched_segments_generator( features, - segments_metadata, + chunks_metadata, batch_size, batched_options, log_progress, @@ -540,14 +462,14 @@ class BatchedInferencePipeline: return segments, info def _batched_segments_generator( - self, features, segments_metadata, batch_size, options, log_progress + self, features, chunks_metadata, batch_size, options, log_progress ): pbar = tqdm(total=len(features), disable=not log_progress, position=0) seg_idx = 0 for i in range(0, len(features), batch_size): results = self.forward( features[i : i + batch_size], - segments_metadata[i : i + batch_size], + chunks_metadata[i : i + batch_size], **options._asdict(), ) @@ -850,7 +772,8 @@ class WhisperModel: elif isinstance(vad_parameters, dict): vad_parameters = VadOptions(**vad_parameters) speech_chunks = get_speech_timestamps(audio, vad_parameters) - audio = collect_chunks(audio, speech_chunks) + audio_chunks, chunks_metadata = collect_chunks(audio, speech_chunks) + audio = torch.cat(audio_chunks, dim=0) duration_after_vad = audio.shape[0] / sampling_rate self.logger.info( @@ -1905,7 +1828,8 @@ class WhisperModel: # get chunks of audio that contain speech speech_chunks = get_speech_timestamps(audio, vad_params) # merge chunks of audio that contain speech into a single array - audio = collect_chunks(audio, speech_chunks) + audio_chunks, chunks_metadata = collect_chunks(audio, speech_chunks) + audio = torch.cat(audio_chunks, dim=0) # calculate new duration of audio without silence duration_vad = audio.shape[0] / sampling_rate diff --git a/faster_whisper/vad.py b/faster_whisper/vad.py index 3881fd8..d448f5b 100644 --- a/faster_whisper/vad.py +++ b/faster_whisper/vad.py @@ -2,18 +2,11 @@ import bisect import functools import os -from abc import ABC -from collections.abc import Callable -from typing import List, NamedTuple, Optional, Union +from typing import Dict, List, NamedTuple, Optional, Tuple import numpy as np import torch -from pyannote.audio.core.io import AudioFile -from pyannote.audio.pipelines import VoiceActivityDetection -from pyannote.audio.pipelines.utils import PipelineModel -from pyannote.core import Annotation, Segment, SlidingWindowFeature - from faster_whisper.utils import get_assets_path @@ -22,9 +15,14 @@ class VadOptions(NamedTuple): """VAD options. Attributes: - threshold: Speech threshold. Silero VAD outputs speech probabilities for each audio chunk, + onset: Speech threshold. Silero VAD outputs speech probabilities for each audio chunk, probabilities ABOVE this value are considered as SPEECH. It is better to tune this parameter for each dataset separately, but "lazy" 0.5 is pretty good for most datasets. + offset: Silence threshold for determining the end of speech. If a probability is lower than + the offset, it is always considered silence. Values higher than offset are only considered + speech if the previous sample was classified as speech; otherwise, they are treated as + silence. This parameter helps refine the detection of speech transitions, ensuring smoother + segment boundaries. min_speech_duration_ms: Final speech chunks shorter min_speech_duration_ms are thrown out. max_speech_duration_s: Maximum duration of speech chunks in seconds. Chunks longer than max_speech_duration_s will be split at the timestamp of the last silence that @@ -35,8 +33,9 @@ class VadOptions(NamedTuple): speech_pad_ms: Final speech chunks are padded by speech_pad_ms each side """ - threshold: float = 0.5 - min_speech_duration_ms: int = 250 + onset: float = 0.5 + offset: float = onset - 0.15 + min_speech_duration_ms: int = 0 max_speech_duration_s: float = float("inf") min_silence_duration_ms: int = 2000 speech_pad_ms: int = 400 @@ -45,6 +44,7 @@ class VadOptions(NamedTuple): def get_speech_timestamps( audio: torch.Tensor, vad_options: Optional[VadOptions] = None, + sampling_rate: int = 16000, **kwargs, ) -> List[dict]: """This method is used for splitting long audios into speech chunks using silero VAD. @@ -52,6 +52,7 @@ def get_speech_timestamps( Args: audio: One dimensional float array. vad_options: Options for VAD processing. + sampling rate: Sampling rate of the audio. kwargs: VAD options passed as keyword arguments for backward compatibility. Returns: @@ -60,13 +61,12 @@ def get_speech_timestamps( if vad_options is None: vad_options = VadOptions(**kwargs) - threshold = vad_options.threshold + onset = vad_options.onset min_speech_duration_ms = vad_options.min_speech_duration_ms max_speech_duration_s = vad_options.max_speech_duration_s min_silence_duration_ms = vad_options.min_silence_duration_ms window_size_samples = 512 speech_pad_ms = vad_options.speech_pad_ms - sampling_rate = 16000 min_speech_samples = sampling_rate * min_speech_duration_ms / 1000 speech_pad_samples = sampling_rate * speech_pad_ms / 1000 max_speech_samples = ( @@ -80,20 +80,16 @@ def get_speech_timestamps( audio_length_samples = len(audio) model = get_vad_model() - state, context = model.get_initial_states(batch_size=1) - speech_probs = [] - for current_start_sample in range(0, audio_length_samples, window_size_samples): - chunk = audio[current_start_sample : current_start_sample + window_size_samples] - if len(chunk) < window_size_samples: - chunk = np.pad(chunk, (0, int(window_size_samples - len(chunk)))) - speech_prob, state, context = model(chunk, state, context, sampling_rate) - speech_probs.append(speech_prob) + padded_audio = np.pad( + audio.numpy(), (0, window_size_samples - audio.shape[0] % window_size_samples) + ) + speech_probs = model(padded_audio.reshape(1, -1)).squeeze(0) triggered = False speeches = [] current_speech = {} - neg_threshold = threshold - 0.15 + offset = vad_options.offset # to save potential segment end (and tolerate some silence) temp_end = 0 @@ -101,12 +97,12 @@ def get_speech_timestamps( prev_end = next_start = 0 for i, speech_prob in enumerate(speech_probs): - if (speech_prob >= threshold) and temp_end: + if (speech_prob >= onset) and temp_end: temp_end = 0 if next_start < prev_end: next_start = window_size_samples * i - if (speech_prob >= threshold) and not triggered: + if (speech_prob >= onset) and not triggered: triggered = True current_speech["start"] = window_size_samples * i continue @@ -133,7 +129,7 @@ def get_speech_timestamps( triggered = False continue - if (speech_prob < neg_threshold) and triggered: + if (speech_prob < offset) and triggered: if not temp_end: temp_end = window_size_samples * i # condition to avoid cutting in very short silence @@ -184,12 +180,27 @@ def get_speech_timestamps( return speeches -def collect_chunks(audio: torch.Tensor, chunks: List[dict]) -> torch.Tensor: - """Collects and concatenates audio chunks.""" +def collect_chunks( + audio: torch.Tensor, chunks: List[dict], sampling_rate: int = 16000 +) -> Tuple[List[torch.Tensor], List[Dict[str, int]]]: + """Collects audio chunks.""" if not chunks: - return torch.tensor([], dtype=torch.float32) + chunk_metadata = { + "start_time": 0, + "end_time": 0, + } + return [torch.tensor([], dtype=torch.float32)], [chunk_metadata] - return torch.cat([audio[chunk["start"] : chunk["end"]] for chunk in chunks]) + audio_chunks = [] + chunks_metadata = [] + for chunk in chunks: + chunk_metadata = { + "start_time": chunk["start"] / sampling_rate, + "end_time": chunk["end"] / sampling_rate, + } + audio_chunks.append(audio[chunk["start"] : chunk["end"]]) + chunks_metadata.append(chunk_metadata) + return audio_chunks, chunks_metadata class SpeechTimestampsMap: @@ -233,12 +244,13 @@ class SpeechTimestampsMap: @functools.lru_cache def get_vad_model(): """Returns the VAD model instance.""" - path = os.path.join(get_assets_path(), "silero_vad.onnx") - return SileroVADModel(path) + encoder_path = os.path.join(get_assets_path(), "silero_encoder_v5.onnx") + decoder_path = os.path.join(get_assets_path(), "silero_decoder_v5.onnx") + return SileroVADModel(encoder_path, decoder_path) class SileroVADModel: - def __init__(self, path): + def __init__(self, encoder_path, decoder_path): try: import onnxruntime except ImportError as e: @@ -247,331 +259,84 @@ class SileroVADModel: ) from e opts = onnxruntime.SessionOptions() - opts.inter_op_num_threads = 1 - opts.intra_op_num_threads = 1 + opts.inter_op_num_threads = 0 + opts.intra_op_num_threads = 0 opts.log_severity_level = 4 - self.session = onnxruntime.InferenceSession( - path, + self.encoder_session = onnxruntime.InferenceSession( + encoder_path, + providers=["CPUExecutionProvider"], + sess_options=opts, + ) + self.decoder_session = onnxruntime.InferenceSession( + decoder_path, providers=["CPUExecutionProvider"], sess_options=opts, ) - def get_initial_states(self, batch_size: int): - state = np.zeros((2, batch_size, 128), dtype=np.float32) - context = np.zeros((batch_size, 64), dtype=np.float32) - return state, context - - def __call__(self, x, state, context, sr: int): - if len(x.shape) == 1: - x = np.expand_dims(x, 0) - if len(x.shape) > 2: - raise ValueError( - f"Too many dimensions for input audio chunk {len(x.shape)}" - ) - if sr / x.shape[1] > 31.25: - raise ValueError("Input audio chunk is too short") - - x = np.concatenate([context, x], axis=1) - - ort_inputs = { - "input": x, - "state": state, - "sr": np.array(sr, dtype="int64"), - } - - out, state = self.session.run(None, ort_inputs) - context = x[..., -64:] - - return out, state, context - - -# BSD 2-Clause License - -# Copyright (c) 2024, Max Bain - -# Redistribution and use in source and binary forms, with or without -# modification, are permitted provided that the following conditions are met: - -# 1. Redistributions of source code must retain the above copyright notice, this -# list of conditions and the following disclaimer. - -# 2. Redistributions in binary form must reproduce the above copyright notice, -# this list of conditions and the following disclaimer in the documentation -# and/or other materials provided with the distribution. - -# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" -# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE -# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE -# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE -# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL -# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR -# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER -# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, -# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - - -# The code below is copied from whisper-x (https://github.com/m-bain/whisperX) -# and adapted for faster_whisper. -class SegmentX: - def __init__(self, start, end, speaker=None): - self.start = start - self.end = end - self.speaker = speaker - - -class VoiceActivitySegmentation(VoiceActivityDetection, ABC): - """Pipeline wrapper class for Voice Activity Segmentation based on VAD scores.""" - - def __init__( - self, - segmentation: PipelineModel = "pyannote/segmentation", - device: Optional[Union[str, torch.device]] = None, - fscore: bool = False, - use_auth_token: Optional[str] = None, - **inference_kwargs, + def __call__( + self, audio: np.ndarray, num_samples: int = 512, context_size_samples: int = 64 ): - """Initialize the pipeline with the model name and the optional device. + assert ( + audio.ndim == 2 + ), "Input should be a 2D tensor with size (batch_size, num_samples)" + assert ( + audio.shape[1] % num_samples == 0 + ), "Input size should be a multiple of num_samples" - Args: - dict parameters of VoiceActivityDetection class from pyannote: - segmentation (PipelineModel): Loaded model name. - device (torch.device or None): Device to perform the segmentation. - fscore (bool): Flag indicating whether to compute F-score during inference. - use_auth_token (str or None): Optional authentication token for model access. - inference_kwargs (dict): Additional arguments from VoiceActivityDetection pipeline. - """ - super().__init__( - segmentation=segmentation, - device=device, - fscore=fscore, - use_auth_token=use_auth_token, - **inference_kwargs, + batch_size = audio.shape[0] + + state = np.zeros((2, batch_size, 128), dtype="float32") + context = np.zeros( + (batch_size, context_size_samples), + dtype="float32", ) - def apply( - self, file: AudioFile, hook: Optional[Callable] = None - ) -> SlidingWindowFeature: - """Apply voice activity detection on the audio file. + batched_audio = audio.reshape(batch_size, -1, num_samples) + context = batched_audio[..., -context_size_samples:] + context[:, -1] = 0 + context = np.roll(context, 1, 1) + batched_audio = np.concatenate([context, batched_audio], 2) - Args: - file (AudioFile): Processed file. - hook (callable): Hook called with signature: hook("step_name", step_artefact, file=file) + batched_audio = batched_audio.reshape(-1, num_samples + context_size_samples) - Returns: - segmentations (SlidingWindowFeature): Voice activity segmentation. - """ - # setup hook (e.g. for debugging purposes) - hook = self.setup_hook(file, hook=hook) + encoder_output = self.encoder_session.run(None, {"input": batched_audio})[0] + encoder_output = encoder_output.reshape(batch_size, -1, 128) - # apply segmentation model if needed - # output shape is (num_chunks, num_frames, 1) - if self.training: - if self.CACHED_SEGMENTATION in file: - segmentations = file[self.CACHED_SEGMENTATION] - else: - segmentations = self._segmentation(file) - file[self.CACHED_SEGMENTATION] = segmentations - else: - segmentations: SlidingWindowFeature = self._segmentation(file) - - return segmentations - - -class BinarizeVadScores: - """Binarize detection scores using hysteresis thresholding. - - Reference: - Gregory Gelly and Jean-Luc Gauvain. "Minimum Word Error Training of - RNN-based Voice Activity Detection", InterSpeech 2015. - - Modified by Max Bain to include WhisperX's min-cut operation - https://arxiv.org/abs/2303.00747 - - """ - - def __init__( - self, - onset: float = 0.5, - offset: Optional[float] = None, - min_duration_on: float = 0.0, - min_duration_off: float = 0.0, - pad_onset: float = 0.0, - pad_offset: float = 0.0, - max_duration: float = float("inf"), - ): - """Initializes the parameters for Binarizing the VAD scores. - - Args: - onset (float, optional): - Onset threshold. Defaults to 0.5. - offset (float, optional): - Offset threshold. Defaults to `onset`. - min_duration_on (float, optional): - Remove active regions shorter than that many seconds. Defaults to 0s. - min_duration_off (float, optional): - Fill inactive regions shorter than that many seconds. Defaults to 0s. - pad_onset (float, optional): - Extend active regions by moving their start time by that many seconds. - Defaults to 0s. - pad_offset (float, optional): - Extend active regions by moving their end time by that many seconds. - Defaults to 0s. - max_duration (float): - The maximum length of an active segment. - """ - super().__init__() - - self.onset = onset - self.offset = offset or onset - - self.pad_onset = pad_onset - self.pad_offset = pad_offset - - self.min_duration_on = min_duration_on - self.min_duration_off = min_duration_off - - self.max_duration = max_duration - - def __get_active_regions(self, scores: SlidingWindowFeature) -> Annotation: - """Extract active regions from VAD scores. - - Args: - scores (SlidingWindowFeature): Detection scores. - - Returns: - active (Annotation): Active regions. - """ - num_frames, num_classes = scores.data.shape - frames = scores.sliding_window - timestamps = [frames[i].middle for i in range(num_frames)] - # annotation meant to store 'active' regions - active = Annotation() - for k, k_scores in enumerate(scores.data.T): - label = k if scores.labels is None else scores.labels[k] - - # initial state - start = timestamps[0] - is_active = k_scores[0] > self.onset - curr_scores = [k_scores[0]] - curr_timestamps = [start] - t = start - # optionally add `strict=False` for python 3.10 or later - for t, y in zip(timestamps[1:], k_scores[1:]): - # currently active - if is_active: - curr_duration = t - start - if curr_duration > self.max_duration: - search_after = len(curr_scores) // 2 - # divide segment - min_score_div_idx = search_after + np.argmin( - curr_scores[search_after:] - ) - min_score_t = curr_timestamps[min_score_div_idx] - region = Segment( - start - self.pad_onset, min_score_t + self.pad_offset - ) - active[region, k] = label - start = curr_timestamps[min_score_div_idx] - curr_scores = curr_scores[min_score_div_idx + 1 :] - curr_timestamps = curr_timestamps[min_score_div_idx + 1 :] - # switching from active to inactive - elif y < self.offset: - region = Segment(start - self.pad_onset, t + self.pad_offset) - active[region, k] = label - start = t - is_active = False - curr_scores = [] - curr_timestamps = [] - curr_scores.append(y) - curr_timestamps.append(t) - # currently inactive - else: - # switching from inactive to active - if y > self.onset: - start = t - is_active = True - - # if active at the end, add final region - if is_active: - region = Segment(start - self.pad_onset, t + self.pad_offset) - active[region, k] = label - - return active - - def __call__(self, scores: SlidingWindowFeature) -> Annotation: - """Binarize detection scores. - - Args: - scores (SlidingWindowFeature): Detection scores. - - Returns: - active (Annotation): Binarized scores. - """ - active = self.__get_active_regions(scores) - # because of padding, some active regions might be overlapping: merge them. - # also: fill same speaker gaps shorter than min_duration_off - if self.pad_offset > 0.0 or self.pad_onset > 0.0 or self.min_duration_off > 0.0: - if self.max_duration < float("inf"): - raise NotImplementedError("This would break current max_duration param") - active = active.support(collar=self.min_duration_off) - - # remove tracks shorter than min_duration_on - if self.min_duration_on > 0: - for segment, track in list(active.itertracks()): - if segment.duration < self.min_duration_on: - del active[segment, track] - - return active - - -def merge_chunks( - segments, - chunk_length, - onset: float = 0.5, - offset: Optional[float] = None, - edge_padding: float = 0.1, -): - """ - Merge operation described in whisper-x paper - """ - curr_end = 0 - merged_segments = [] - seg_idxs = [] - speaker_idxs = [] - - assert chunk_length > 0 - binarize = BinarizeVadScores(max_duration=chunk_length, onset=onset, offset=offset) - segments = binarize(segments) - segments_list = [] - for speech_turn in segments.get_timeline(): - segments_list.append( - SegmentX( - max(0.0, speech_turn.start - edge_padding), - speech_turn.end + edge_padding, - "UNKNOWN", + decoder_outputs = [] + for window in np.split(encoder_output, encoder_output.shape[1], axis=1): + out, state = self.decoder_session.run( + None, {"input": window.squeeze(1), "state": state} ) - ) # 100ms edge padding to account for edge errors + decoder_outputs.append(out) - if len(segments_list) == 0: - print("No active speech found in audio") + out = np.stack(decoder_outputs, axis=1).squeeze(-1) + return out + + +def merge_segments(segments_list, vad_options: VadOptions, sampling_rate: int = 16000): + if not segments_list: return [] - # Make sur the starting point is the start of the segment. - curr_start = segments_list[0].start + curr_end = 0 + seg_idxs = [] + merged_segments = [] + edge_padding = vad_options.speech_pad_ms * sampling_rate // 1000 + chunk_length = vad_options.max_speech_duration_s * sampling_rate + + curr_start = segments_list[0]["start"] for idx, seg in enumerate(segments_list): # if any segment start timing is less than previous segment end timing, # reset the edge padding. Similarly for end timing. if idx > 0: - if seg.start < segments_list[idx - 1].end: - seg.start += edge_padding + if seg["start"] < segments_list[idx - 1]["end"]: + seg["start"] += edge_padding if idx < len(segments_list) - 1: - if seg.end > segments_list[idx + 1].start: - seg.end -= edge_padding + if seg["end"] > segments_list[idx + 1]["start"]: + seg["end"] -= edge_padding - if seg.end - curr_start > chunk_length and curr_end - curr_start > 0: + if seg["end"] - curr_start > chunk_length and curr_end - curr_start > 0: merged_segments.append( { "start": curr_start, @@ -579,12 +344,10 @@ def merge_chunks( "segments": seg_idxs, } ) - curr_start = seg.start + curr_start = seg["start"] seg_idxs = [] - speaker_idxs = [] - curr_end = seg.end - seg_idxs.append((seg.start, seg.end)) - speaker_idxs.append(seg.speaker) + curr_end = seg["end"] + seg_idxs.append((seg["start"], seg["end"])) # add final merged_segments.append( { diff --git a/requirements.txt b/requirements.txt index 699d160..71fc482 100644 --- a/requirements.txt +++ b/requirements.txt @@ -2,7 +2,6 @@ ctranslate2>=4.0,<5 huggingface_hub>=0.13 tokenizers>=0.13,<1 onnxruntime>=1.14,<2 -pyannote-audio>=3.1.1 torch>=2.1.1 av>=11 tqdm \ No newline at end of file diff --git a/tests/test_transcribe.py b/tests/test_transcribe.py index 96eb68c..08cc3cc 100644 --- a/tests/test_transcribe.py +++ b/tests/test_transcribe.py @@ -39,8 +39,10 @@ def test_transcribe(jfk_path): assert segment.text == "".join(word.word for word in segment.words) assert segment.start == segment.words[0].start assert segment.end == segment.words[-1].end - batched_model = BatchedInferencePipeline(model=model, use_vad_model=False) - result, info = batched_model.transcribe(jfk_path, word_timestamps=True) + batched_model = BatchedInferencePipeline(model=model) + result, info = batched_model.transcribe( + jfk_path, word_timestamps=True, vad_filter=False + ) assert info.language == "en" assert info.language_probability > 0.7 segments = [] @@ -68,7 +70,7 @@ def test_batched_transcribe(physcisworks_path): {"start": segment.start, "end": segment.end, "text": segment.text} ) # number of near 30 sec segments - assert len(segments) == 8 + assert len(segments) == 7 result, info = batched_model.transcribe( physcisworks_path, @@ -82,7 +84,7 @@ def test_batched_transcribe(physcisworks_path): segments.append( {"start": segment.start, "end": segment.end, "text": segment.text} ) - assert len(segments) > 8 + assert len(segments) > 7 def test_prefix_with_timestamps(jfk_path):