Voice Activity Detection (VAD)

1# Voice Activity Detection (VAD)
2 
3## Overview
4 
5Voice Activity Detection identifies which parts of an audio signal contain speech versus silence or background noise. This is a critical first step in speaker diarization pipelines.
6 
7## When to Use
8 
9- Preprocessing audio before speaker diarization
10- Filtering out silence and noise
11- Segmenting audio into speech chunks
12- Improving diarization accuracy by focusing on speech regions
13 
14## Available VAD Tools
15 
16### 1. Silero VAD (Recommended for Short Segments)
17 
18**Best for:** Short audio segments, real-time applications, better detection of brief speech
19 
20```python
21import torch
22 
23# Load Silero VAD model
24model, utils = torch.hub.load(
25    repo_or_dir='snakers4/silero-vad',
26    model='silero_vad',
27    force_reload=False,
28    onnx=False
29)
30get_speech_timestamps = utils[0]
31 
32# Run VAD
33speech_timestamps = get_speech_timestamps(
34    waveform[0],  # mono audio waveform
35    model,
36    threshold=0.6,  # speech probability threshold
37    min_speech_duration_ms=350,  # minimum speech segment length
38    min_silence_duration_ms=400,  # minimum silence between segments
39    sampling_rate=sample_rate
40)
41 
42# Convert to boundaries format
43boundaries = [[ts['start'] / sample_rate, ts['end'] / sample_rate]
44              for ts in speech_timestamps]
45```
46 
47**Advantages:**
48- Better at detecting short speech segments
49- Lower false alarm rate
50- Optimized for real-time processing
51 
52### 2. SpeechBrain VAD
53 
54**Best for:** General-purpose VAD, longer audio files
55 
56```python
57…