silero_vad_torch_hub_speech_timestamp_detection.py

1import torch
2torch.set_num_threads(1)
3
4# Load the model and utils locally or from torch.hub
5model, utils = torch.hub.load(repo_or_dir='snakers4/silero-vad',
6                              model='silero_vad',
7                              force_reload=True,
8                              onnx=False)
9
10(get_speech_timestamps,
11 save_audio,
12 read_audio,
13 VADIterator,
14 collect_chunks) = utils
15
16# Load audio (replace 'test.wav' with your audio file path)
17# Sampling rate should be 8000 or 16000
18SAMPLING_RATE = 16000
19wav = read_audio('test.wav', sampling_rate=SAMPLING_RATE)
20
21# Get speech timestamps from entire audio file
22speech_timestamps = get_speech_timestamps(wav, model, sampling_rate=SAMPLING_RATE)
23print(speech_timestamps)
24
25# Merge all speech chunks into one audio file
26save_audio('only_speech.wav',
27           collect_chunks(speech_timestamps, wav), sampling_rate=SAMPLING_RATE)