import torch import torchaudio from typing import Callable, List import warnings from packaging import version languages = ['ru', 'en', 'de', 'es'] class OnnxWrapper(): def __init__(self, path, force_onnx_cpu=False): import numpy as np global np import onnxruntime opts = onnxruntime.SessionOptions() opts.inter_op_num_threads = 1 opts.intra_op_num_threads = 1 if force_onnx_cpu and 'CPUExecutionProvider' in onnxruntime.get_available_providers(): self.session = onnxruntime.InferenceSession(path, providers=['CPUExecutionProvider'], sess_options=opts) else: self.session = onnxruntime.InferenceSession(path, sess_options=opts) self.reset_states() if '16k' in path: warnings.warn('This model support only 16000 sampling rate!') self.sample_rates = [16000] else: self.sample_rates = [8000, 16000] def _validate_input(self, x, sr: int): if x.dim() == 1: x = x.unsqueeze(0) if x.dim() > 2: raise ValueError(f"Too many dimensions for input audio chunk {x.dim()}") if sr != 16000 and (sr % 16000 == 0): step = sr // 16000 x = x[:,::step] sr = 16000 if sr not in self.sample_rates: raise ValueError(f"Supported sampling rates: {self.sample_rates} (or multiply of 16000)") if sr / x.shape[1] > 31.25: raise ValueError("Input audio chunk is too short") return x, sr def reset_states(self, batch_size=1): self._state = torch.zeros((2, batch_size, 128)).float() self._context = torch.zeros(0) self._last_sr = 0 self._last_batch_size = 0 def __call__(self, x, sr: int): x, sr = self._validate_input(x, sr) num_samples = 512 if sr == 16000 else 256 if x.shape[-1] != num_samples: raise ValueError(f"Provided number of samples is {x.shape[-1]} (Supported values: 256 for 8000 sample rate, 512 for 16000)") batch_size = x.shape[0] context_size = 64 if sr == 16000 else 32 if not self._last_batch_size: self.reset_states(batch_size) if (self._last_sr) and (self._last_sr != sr): self.reset_states(batch_size) if (self._last_batch_size) and (self._last_batch_size != batch_size): self.reset_states(batch_size) if not len(self._context): self._context = torch.zeros(batch_size, context_size) x = torch.cat([self._context, x], dim=1) if sr in [8000, 16000]: ort_inputs = {'input': x.numpy(), 'state': self._state.numpy(), 'sr': np.array(sr, dtype='int64')} ort_outs = self.session.run(None, ort_inputs) out, state = ort_outs self._state = torch.from_numpy(state) else: raise ValueError() self._context = x[..., -context_size:] self._last_sr = sr self._last_batch_size = batch_size out = torch.from_numpy(out) return out def audio_forward(self, x, sr: int): outs = [] x, sr = self._validate_input(x, sr) self.reset_states() num_samples = 512 if sr == 16000 else 256 if x.shape[1] % num_samples: pad_num = num_samples - (x.shape[1] % num_samples) x = torch.nn.functional.pad(x, (0, pad_num), 'constant', value=0.0) for i in range(0, x.shape[1], num_samples): wavs_batch = x[:, i:i+num_samples] out_chunk = self.__call__(wavs_batch, sr) outs.append(out_chunk) stacked = torch.cat(outs, dim=1) return stacked.cpu() class Validator(): def __init__(self, url, force_onnx_cpu): self.onnx = True if url.endswith('.onnx') else False torch.hub.download_url_to_file(url, 'inf.model') if self.onnx: import onnxruntime if force_onnx_cpu and 'CPUExecutionProvider' in onnxruntime.get_available_providers(): self.model = onnxruntime.InferenceSession('inf.model', providers=['CPUExecutionProvider']) else: self.model = onnxruntime.InferenceSession('inf.model') else: self.model = init_jit_model(model_path='inf.model') def __call__(self, inputs: torch.Tensor): with torch.no_grad(): if self.onnx: ort_inputs = {'input': inputs.cpu().numpy()} outs = self.model.run(None, ort_inputs) outs = [torch.Tensor(x) for x in outs] else: outs = self.model(inputs) return outs def read_audio(path: str, sampling_rate: int = 16000) -> torch.Tensor: ta_ver = version.parse(torchaudio.__version__) if ta_ver < version.parse("2.9"): try: effects = [['channels', '1'],['rate', str(sampling_rate)]] wav, sr = torchaudio.sox_effects.apply_effects_file(path, effects=effects) except: wav, sr = torchaudio.load(path) else: try: wav, sr = torchaudio.load(path) except: try: from torchcodec.decoders import AudioDecoder samples = AudioDecoder(path).get_all_samples() wav = samples.data sr = samples.sample_rate except ImportError: raise RuntimeError( f"torchaudio version {torchaudio.__version__} requires torchcodec for audio I/O. " + "Install torchcodec or pin torchaudio < 2.9" ) if wav.ndim > 1 and wav.size(0) > 1: wav = wav.mean(dim=0, keepdim=True) if sr != sampling_rate: wav = torchaudio.transforms.Resample(sr, sampling_rate)(wav) return wav.squeeze(0) def save_audio(path: str, tensor: torch.Tensor, sampling_rate: int = 16000): tensor = tensor.detach().cpu() if tensor.ndim == 1: tensor = tensor.unsqueeze(0) ta_ver = version.parse(torchaudio.__version__) try: torchaudio.save(path, tensor, sampling_rate, bits_per_sample=16) except Exception: if ta_ver >= version.parse("2.9"): try: from torchcodec.encoders import AudioEncoder encoder = AudioEncoder(tensor, sample_rate=16000) encoder.to_file(path) except ImportError: raise RuntimeError( f"torchaudio version {torchaudio.__version__} requires torchcodec for saving. " + "Install torchcodec or pin torchaudio < 2.9" ) else: raise def init_jit_model(model_path: str, device=torch.device('cpu')): model = torch.jit.load(model_path, map_location=device) model.eval() return model def make_visualization(probs, step): import pandas as pd pd.DataFrame({'probs': probs}, index=[x * step for x in range(len(probs))]).plot(figsize=(16, 8), kind='area', ylim=[0, 1.05], xlim=[0, len(probs) * step], xlabel='seconds', ylabel='speech probability', colormap='tab20') @torch.no_grad() def get_speech_timestamps(audio: torch.Tensor, model, threshold: float = 0.5, sampling_rate: int = 16000, min_speech_duration_ms: int = 250, max_speech_duration_s: float = float('inf'), min_silence_duration_ms: int = 100, speech_pad_ms: int = 30, return_seconds: bool = False, time_resolution: int = 1, visualize_probs: bool = False, progress_tracking_callback: Callable[[float], None] = None, neg_threshold: float = None, window_size_samples: int = 512, min_silence_at_max_speech: int = 98, use_max_poss_sil_at_max_speech: bool = True): """ This method is used for splitting long audios into speech chunks using silero VAD Parameters ---------- audio: torch.Tensor, one dimensional One dimensional float torch.Tensor, other types are casted to torch if possible model: preloaded .jit/.onnx silero VAD model threshold: float (default - 0.5) 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. sampling_rate: int (default - 16000) Currently silero VAD models support 8000 and 16000 (or multiply of 16000) sample rates min_speech_duration_ms: int (default - 250 milliseconds) Final speech chunks shorter min_speech_duration_ms are thrown out max_speech_duration_s: int (default - inf) 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 lasts more than 100ms (if any), to prevent aggressive cutting. Otherwise, they will be split aggressively just before max_speech_duration_s. min_silence_duration_ms: int (default - 100 milliseconds) In the end of each speech chunk wait for min_silence_duration_ms before separating it speech_pad_ms: int (default - 30 milliseconds) Final speech chunks are padded by speech_pad_ms each side return_seconds: bool (default - False) whether return timestamps in seconds (default - samples) time_resolution: bool (default - 1) time resolution of speech coordinates when requested as seconds visualize_probs: bool (default - False) whether draw prob hist or not progress_tracking_callback: Callable[[float], None] (default - None) callback function taking progress in percents as an argument neg_threshold: float (default = threshold - 0.15) Negative threshold (noise or exit threshold). If model's current state is SPEECH, values BELOW this value are considered as NON-SPEECH. min_silence_at_max_speech: int (default - 98ms) Minimum silence duration in ms which is used to avoid abrupt cuts when max_speech_duration_s is reached use_max_poss_sil_at_max_speech: bool (default - True) Whether to use the maximum possible silence at max_speech_duration_s or not. If not, the last silence is used. window_size_samples: int (default - 512 samples) !!! DEPRECATED, DOES NOTHING !!! Returns ---------- speeches: list of dicts list containing ends and beginnings of speech chunks (samples or seconds based on return_seconds) """ if not torch.is_tensor(audio): try: audio = torch.Tensor(audio) except: raise TypeError("Audio cannot be casted to tensor. Cast it manually") if len(audio.shape) > 1: for i in range(len(audio.shape)): # trying to squeeze empty dimensions audio = audio.squeeze(0) if len(audio.shape) > 1: raise ValueError("More than one dimension in audio. Are you trying to process audio with 2 channels?") if sampling_rate > 16000 and (sampling_rate % 16000 == 0): step = sampling_rate // 16000 sampling_rate = 16000 audio = audio[::step] warnings.warn('Sampling rate is a multiply of 16000, casting to 16000 manually!') else: step = 1 if sampling_rate not in [8000, 16000]: raise ValueError("Currently silero VAD models support 8000 and 16000 (or multiply of 16000) sample rates") window_size_samples = 512 if sampling_rate == 16000 else 256 model.reset_states() min_speech_samples = sampling_rate * min_speech_duration_ms / 1000 speech_pad_samples = sampling_rate * speech_pad_ms / 1000 max_speech_samples = sampling_rate * max_speech_duration_s - window_size_samples - 2 * speech_pad_samples min_silence_samples = sampling_rate * min_silence_duration_ms / 1000 min_silence_samples_at_max_speech = sampling_rate * min_silence_at_max_speech / 1000 audio_length_samples = len(audio) 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 = torch.nn.functional.pad(chunk, (0, int(window_size_samples - len(chunk)))) speech_prob = model(chunk, sampling_rate).item() speech_probs.append(speech_prob) # calculate progress and send it to callback function progress = current_start_sample + window_size_samples if progress > audio_length_samples: progress = audio_length_samples progress_percent = (progress / audio_length_samples) * 100 if progress_tracking_callback: progress_tracking_callback(progress_percent) triggered = False speeches = [] current_speech = {} if neg_threshold is None: neg_threshold = max(threshold - 0.15, 0.01) temp_end = 0 # to save potential segment end (and tolerate some silence) prev_end = next_start = 0 # to save potential segment limits in case of maximum segment size reached possible_ends = [] for i, speech_prob in enumerate(speech_probs): cur_sample = window_size_samples * i # If speech returns after a temp_end, record candidate silence if long enough and clear temp_end if (speech_prob >= threshold) and temp_end: sil_dur = cur_sample - temp_end if sil_dur > min_silence_samples_at_max_speech: possible_ends.append((temp_end, sil_dur)) temp_end = 0 if next_start < prev_end: next_start = cur_sample # Start of speech if (speech_prob >= threshold) and not triggered: triggered = True current_speech['start'] = cur_sample continue # Max speech length reached: decide where to cut if triggered and (cur_sample - current_speech['start'] > max_speech_samples): if use_max_poss_sil_at_max_speech and possible_ends: prev_end, dur = max(possible_ends, key=lambda x: x[1]) # use the longest possible silence segment in the current speech chunk current_speech['end'] = prev_end speeches.append(current_speech) current_speech = {} next_start = prev_end + dur if next_start < prev_end + cur_sample: # previously reached silence (< neg_thres) and is still not speech (< thres) current_speech['start'] = next_start else: triggered = False prev_end = next_start = temp_end = 0 possible_ends = [] else: # Legacy max-speech cut (use_max_poss_sil_at_max_speech=False): prefer last valid silence (prev_end) if available if prev_end: current_speech['end'] = prev_end speeches.append(current_speech) current_speech = {} if next_start < prev_end: triggered = False else: current_speech['start'] = next_start prev_end = next_start = temp_end = 0 possible_ends = [] else: # No prev_end -> fallback to cutting at current sample current_speech['end'] = cur_sample speeches.append(current_speech) current_speech = {} prev_end = next_start = temp_end = 0 triggered = False possible_ends = [] continue # Silence detection while in speech if (speech_prob < neg_threshold) and triggered: if not temp_end: temp_end = cur_sample sil_dur_now = cur_sample - temp_end if not use_max_poss_sil_at_max_speech and sil_dur_now > min_silence_samples_at_max_speech: # condition to avoid cutting in very short silence prev_end = temp_end if sil_dur_now < min_silence_samples: continue else: current_speech['end'] = temp_end if (current_speech['end'] - current_speech['start']) > min_speech_samples: speeches.append(current_speech) current_speech = {} prev_end = next_start = temp_end = 0 triggered = False possible_ends = [] continue if current_speech and (audio_length_samples - current_speech['start']) > min_speech_samples: current_speech['end'] = audio_length_samples speeches.append(current_speech) for i, speech in enumerate(speeches): if i == 0: speech['start'] = int(max(0, speech['start'] - speech_pad_samples)) if i != len(speeches) - 1: silence_duration = speeches[i+1]['start'] - speech['end'] if silence_duration < 2 * speech_pad_samples: speech['end'] += int(silence_duration // 2) speeches[i+1]['start'] = int(max(0, speeches[i+1]['start'] - silence_duration // 2)) else: speech['end'] = int(min(audio_length_samples, speech['end'] + speech_pad_samples)) speeches[i+1]['start'] = int(max(0, speeches[i+1]['start'] - speech_pad_samples)) else: speech['end'] = int(min(audio_length_samples, speech['end'] + speech_pad_samples)) if return_seconds: audio_length_seconds = audio_length_samples / sampling_rate for speech_dict in speeches: speech_dict['start'] = max(round(speech_dict['start'] / sampling_rate, time_resolution), 0) speech_dict['end'] = min(round(speech_dict['end'] / sampling_rate, time_resolution), audio_length_seconds) elif step > 1: for speech_dict in speeches: speech_dict['start'] *= step speech_dict['end'] *= step if visualize_probs: make_visualization(speech_probs, window_size_samples / sampling_rate) return speeches class VADIterator: def __init__(self, model, threshold: float = 0.5, sampling_rate: int = 16000, min_silence_duration_ms: int = 100, speech_pad_ms: int = 30 ): """ Class for stream imitation Parameters ---------- model: preloaded .jit/.onnx silero VAD model threshold: float (default - 0.5) 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. sampling_rate: int (default - 16000) Currently silero VAD models support 8000 and 16000 sample rates min_silence_duration_ms: int (default - 100 milliseconds) In the end of each speech chunk wait for min_silence_duration_ms before separating it speech_pad_ms: int (default - 30 milliseconds) Final speech chunks are padded by speech_pad_ms each side """ self.model = model self.threshold = threshold self.sampling_rate = sampling_rate if sampling_rate not in [8000, 16000]: raise ValueError('VADIterator does not support sampling rates other than [8000, 16000]') self.min_silence_samples = sampling_rate * min_silence_duration_ms / 1000 self.speech_pad_samples = sampling_rate * speech_pad_ms / 1000 self.reset_states() def reset_states(self): self.model.reset_states() self.triggered = False self.temp_end = 0 self.current_sample = 0 @torch.no_grad() def __call__(self, x, return_seconds=False, time_resolution: int = 1): """ x: torch.Tensor audio chunk (see examples in repo) return_seconds: bool (default - False) whether return timestamps in seconds (default - samples) time_resolution: int (default - 1) time resolution of speech coordinates when requested as seconds """ if not torch.is_tensor(x): try: x = torch.Tensor(x) except: raise TypeError("Audio cannot be casted to tensor. Cast it manually") window_size_samples = len(x[0]) if x.dim() == 2 else len(x) self.current_sample += window_size_samples speech_prob = self.model(x, self.sampling_rate).item() if (speech_prob >= self.threshold) and self.temp_end: self.temp_end = 0 if (speech_prob >= self.threshold) and not self.triggered: self.triggered = True speech_start = max(0, self.current_sample - self.speech_pad_samples - window_size_samples) return {'start': int(speech_start) if not return_seconds else round(speech_start / self.sampling_rate, time_resolution)} if (speech_prob < self.threshold - 0.15) and self.triggered: if not self.temp_end: self.temp_end = self.current_sample if self.current_sample - self.temp_end < self.min_silence_samples: return None else: speech_end = self.temp_end + self.speech_pad_samples - window_size_samples self.temp_end = 0 self.triggered = False return {'end': int(speech_end) if not return_seconds else round(speech_end / self.sampling_rate, time_resolution)} return None def collect_chunks(tss: List[dict], wav: torch.Tensor, seconds: bool = False, sampling_rate: int = None) -> torch.Tensor: """Collect audio chunks from a longer audio clip This method extracts audio chunks from an audio clip, using a list of provided coordinates, and concatenates them together. Coordinates can be passed either as sample numbers or in seconds, in which case the audio sampling rate is also needed. Parameters ---------- tss: List[dict] Coordinate list of the clips to collect from the audio. wav: torch.Tensor, one dimensional One dimensional float torch.Tensor, containing the audio to clip. seconds: bool (default - False) Whether input coordinates are passed as seconds or samples. sampling_rate: int (default - None) Input audio sampling rate. Required if seconds is True. Returns ------- torch.Tensor, one dimensional One dimensional float torch.Tensor of the concatenated clipped audio chunks. Raises ------ ValueError Raised if sampling_rate is not provided when seconds is True. """ if seconds and not sampling_rate: raise ValueError('sampling_rate must be provided when seconds is True') chunks = list() _tss = _seconds_to_samples_tss(tss, sampling_rate) if seconds else tss for i in _tss: chunks.append(wav[i['start']:i['end']]) return torch.cat(chunks) def drop_chunks(tss: List[dict], wav: torch.Tensor, seconds: bool = False, sampling_rate: int = None) -> torch.Tensor: """Drop audio chunks from a longer audio clip This method extracts audio chunks from an audio clip, using a list of provided coordinates, and drops them. Coordinates can be passed either as sample numbers or in seconds, in which case the audio sampling rate is also needed. Parameters ---------- tss: List[dict] Coordinate list of the clips to drop from from the audio. wav: torch.Tensor, one dimensional One dimensional float torch.Tensor, containing the audio to clip. seconds: bool (default - False) Whether input coordinates are passed as seconds or samples. sampling_rate: int (default - None) Input audio sampling rate. Required if seconds is True. Returns ------- torch.Tensor, one dimensional One dimensional float torch.Tensor of the input audio minus the dropped chunks. Raises ------ ValueError Raised if sampling_rate is not provided when seconds is True. """ if seconds and not sampling_rate: raise ValueError('sampling_rate must be provided when seconds is True') chunks = list() cur_start = 0 _tss = _seconds_to_samples_tss(tss, sampling_rate) if seconds else tss for i in _tss: chunks.append((wav[cur_start: i['start']])) cur_start = i['end'] chunks.append(wav[cur_start:]) return torch.cat(chunks) def _seconds_to_samples_tss(tss: List[dict], sampling_rate: int) -> List[dict]: """Convert coordinates expressed in seconds to sample coordinates. """ return [{ 'start': round(crd['start']) * sampling_rate, 'end': round(crd['end']) * sampling_rate } for crd in tss]