WebRTC-VAD

WebRTC-VAD is the core, low-latency algorithm that classifies real-time audio streams into speech or silence for efficient communication protocols.

WebRTC Voice Activity Detection (VAD) is a critical, real-time component for optimizing audio transmission efficiency: it determines if human speech is present. The algorithm operates by segmenting the incoming audio stream into short frames: specifically, 10, 20, or 30ms segments are analyzed. It extracts key features—including energy levels and spectral content—to make a high-speed, binary classification decision (speech or non-speech). This process is vital for Voice over IP (VoIP) applications, enabling Discontinuous Transmission (DTX) to save significant network bandwidth and reducing computational load in systems like Automatic Speech Recognition (ASR). The system maintains high fidelity by using adaptive thresholds and noise suppression to minimize both false positives and false negatives.

https://webrtc.org