import logging import warnings from dataclasses import dataclass from functools import lru_cache import click import numpy as np from huggingface_hub import hf_hub_download from numpy.typing import NDArray from ..utils import AudioChunk, audio_to_float32 from .protocol import PauseDetectionModel logger = logging.getLogger(__name__) # The code below is adapted from https://github.com/snakers4/silero-vad. # The code below is adapted from https://github.com/gpt-omni/mini-omni/blob/main/utils/vad.py @lru_cache def get_silero_model() -> PauseDetectionModel: """Returns the VAD model instance and warms it up with dummy data.""" # Warm up the model with dummy data try: import importlib.util mod = importlib.util.find_spec("onnxruntime") if mod is None: raise RuntimeError("Install fastrtc[vad] to use ReplyOnPause") except (ValueError, ModuleNotFoundError): raise RuntimeError("Install fastrtc[vad] to use ReplyOnPause") model = SileroVADModel() print(click.style("INFO", fg="green") + ":\t Warming up VAD model.") model.warmup() print(click.style("INFO", fg="green") + ":\t VAD model warmed up.") return model @dataclass class SileroVadOptions: """VAD options. Attributes: threshold: 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. 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 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: In the end of each speech chunk wait for min_silence_duration_ms before separating it window_size_samples: Audio chunks of window_size_samples size are fed to the silero VAD model. WARNING! Silero VAD models were trained using 512, 1024, 1536 samples for 16000 sample rate. Values other than these may affect model performance!! speech_pad_ms: Final speech chunks are padded by speech_pad_ms each side speech_duration: If the length of the speech is less than this value, a pause will be detected. """ threshold: float = 0.5 min_speech_duration_ms: int = 250 max_speech_duration_s: float = float("inf") min_silence_duration_ms: int = 2000 window_size_samples: int = 1024 speech_pad_ms: int = 400 class SileroVADModel: @staticmethod def download_model() -> str: return hf_hub_download( repo_id="freddyaboulton/silero-vad", filename="silero_vad.onnx" ) def __init__(self): try: import onnxruntime except ImportError as e: raise RuntimeError( "Applying the VAD filter requires the onnxruntime package" ) from e path = self.download_model() opts = onnxruntime.SessionOptions() opts.inter_op_num_threads = 1 opts.intra_op_num_threads = 1 opts.log_severity_level = 4 self.session = onnxruntime.InferenceSession( path, providers=["CPUExecutionProvider"], sess_options=opts, ) def get_initial_state(self, batch_size: int): h = np.zeros((2, batch_size, 64), dtype=np.float32) c = np.zeros((2, batch_size, 64), dtype=np.float32) return h, c @staticmethod def collect_chunks(audio: np.ndarray, chunks: list[AudioChunk]) -> np.ndarray: """Collects and concatenates audio chunks.""" if not chunks: return np.array([], dtype=np.float32) return np.concatenate( [audio[chunk["start"] : chunk["end"]] for chunk in chunks] ) def get_speech_timestamps( self, audio: np.ndarray, vad_options: SileroVadOptions, **kwargs, ) -> list[AudioChunk]: """This method is used for splitting long audios into speech chunks using silero VAD. Args: audio: One dimensional float array. vad_options: Options for VAD processing. kwargs: VAD options passed as keyword arguments for backward compatibility. Returns: List of dicts containing begin and end samples of each speech chunk. """ threshold = vad_options.threshold 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 = vad_options.window_size_samples speech_pad_ms = vad_options.speech_pad_ms if window_size_samples not in [512, 1024, 1536]: warnings.warn( "Unusual window_size_samples! Supported window_size_samples:\n" " - [512, 1024, 1536] for 16000 sampling_rate" ) 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 = ( 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 * 98 / 1000 audio_length_samples = len(audio) state = self.get_initial_state(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 = self(chunk, state, sampling_rate) speech_probs.append(speech_prob) triggered = False speeches = [] current_speech = {} neg_threshold = threshold - 0.15 # to save potential segment end (and tolerate some silence) temp_end = 0 # to save potential segment limits in case of maximum segment size reached prev_end = next_start = 0 for i, speech_prob in enumerate(speech_probs): if (speech_prob >= threshold) and temp_end: temp_end = 0 if next_start < prev_end: next_start = window_size_samples * i if (speech_prob >= threshold) and not triggered: triggered = True current_speech["start"] = window_size_samples * i continue if ( triggered and (window_size_samples * i) - current_speech["start"] > max_speech_samples ): if prev_end: current_speech["end"] = prev_end speeches.append(current_speech) current_speech = {} # previously reached silence (< neg_thres) and is still not speech (< thres) if next_start < prev_end: triggered = False else: current_speech["start"] = next_start prev_end = next_start = temp_end = 0 else: current_speech["end"] = window_size_samples * i speeches.append(current_speech) current_speech = {} prev_end = next_start = temp_end = 0 triggered = False continue if (speech_prob < neg_threshold) and triggered: if not temp_end: temp_end = window_size_samples * i # condition to avoid cutting in very short silence if ( window_size_samples * i ) - temp_end > min_silence_samples_at_max_speech: prev_end = temp_end if (window_size_samples * i) - temp_end < 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 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) ) return speeches def warmup(self): for _ in range(10): dummy_audio = np.zeros(102400, dtype=np.float32) self.vad((24000, dummy_audio), None) def vad( self, audio: tuple[int, NDArray[np.float32] | NDArray[np.int16]], options: None | SileroVadOptions, ) -> tuple[float, list[AudioChunk]]: sampling_rate, audio_ = audio logger.debug("VAD audio shape input: %s", audio_.shape) try: audio_ = audio_to_float32(audio_) sr = 16000 if sr != sampling_rate: try: import librosa # type: ignore except ImportError as e: raise RuntimeError( "Applying the VAD filter requires the librosa if the input sampling rate is not 16000hz" ) from e audio_ = librosa.resample(audio_, orig_sr=sampling_rate, target_sr=sr) if not options: options = SileroVadOptions() speech_chunks = self.get_speech_timestamps(audio_, options) logger.debug("VAD speech chunks: %s", speech_chunks) audio_ = self.collect_chunks(audio_, speech_chunks) logger.debug("VAD audio shape: %s", audio_.shape) duration_after_vad = audio_.shape[0] / sr return duration_after_vad, speech_chunks except Exception as e: import math import traceback logger.debug("VAD Exception: %s", str(e)) exec = traceback.format_exc() logger.debug("traceback %s", exec) return math.inf, [] def __call__(self, x, state, 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: # type: ignore raise ValueError("Input audio chunk is too short") h, c = state ort_inputs = { "input": x, "h": h, "c": c, "sr": np.array(sr, dtype="int64"), } out, h, c = self.session.run(None, ort_inputs) state = (h, c) return out, state