snakers4/silero-vad

Silero VAD: pre-trained enterprise-grade Voice Activity Detector

What it solves

Silero VAD is a pre-trained, enterprise-grade Voice Activity Detector (VAD) designed to accurately identify segments of audio containing speech versus silence or background noise. It is optimized for high performance and portability across various hardware environments, including edge devices and mobile platforms.

How it works

The project provides a pre-trained model that can be processed via PyTorch or ONNX Runtime. It supports sampling rates of 8000 Hz and 16000 Hz and can process audio chunks (30+ ms) in less than 1ms on a single CPU thread. The model is lightweight (approximately 2MB) and was trained on a massive dataset covering over 6,000 languages to ensure robustness across different domains and noise levels.

Who it’s for

  • Developers of voice interfaces and voice bots: To detect when a user is speaking.
  • IoT and edge device engineers: To implement lightweight, high-performance speech detection on limited hardware.
  • Data scientists: For cleaning and preparing audio datasets by removing non-speech segments.
  • Telephony and call-center automation engineers: To automate voice-based interactions.

Highlights

  • High Accuracy: Delivers stellar results on speech detection tasks.
  • High Performance: Extremely fast processing times (under 1ms per chunk on CPU) with optional GPU and ONNX acceleration.
  • Broad Language Support: Trained on 6,000+ languages for global applicability.
  • Highly Portable: Compatible with PyTorch and ONNX, making it deployable across various runtimes and languages (C++, Rust, Go, Java, C#).
  • Lightweight: Small model size (~2MB) for easy deployment.
  • Permissive License: Released under the MIT license with no telemetry or vendor lock-in.

Related

  • Project
    ricky0123/vad

    A JavaScript library for voice activity detection that allows developers to run callbacks on audio segments containing user speech.

  • Project
    filippogiruzzi/voice_activity_detection

    A deep learning-based Voice Activity Detection system using a 1D-ResNet and MFCC features to classify audio signals as speech or noise.

  • Project
    amsehili/auditok

    A lightweight Python library for audio activity detection that splits audio streams into events using energy thresholding or WebRTC VAD.

  • Project
    modelscope/FunASR

    FunASR is an industrial-grade speech recognition toolkit from Alibaba's ModelScope team. It provides a modular Python API, CLI, and server for offline/streaming ASR, plus VAD, punctuation, speaker diarization, emotion recognition, and edge deployment via llama.cpp. It includes multiple pre-trained models (e.g., Fun-ASR-Nano, SenseVoiceSmall) and supports GPU, CPU, and edge devices.

  • Project
    KoljaB/RealtimeSTT

    A Python speech-to-text library that provides fast, real-time transcription with integrated voice activity detection and wake word support.