silero-go

Lightweight Go wrapper around the Silero Voice Activity Detector using ONNX Runtime and miniaudio.

This library enables real-time or file-based voice activity detection (VAD) in Go. It wraps the Silero VAD model and provides microphone input/output and WAV file integration with native audio processing.

---

Features

• 🧠 ONNX-based Silero VAD inference
• 🎙️ Real-time microphone speech detection
• 📁 WAV file-based speech detection
• 🔊 Microphone and WAV playback support
• 🔄 Optional WAV conversion (resample, format, channels)

---

Requirements

• Go 1.20+
• ONNX Runtime installed on your system
• Silero VAD model .onnx file (see below)

Install ONNX Runtime

This library uses yalue/onnxruntime_go.
Please follow its installation instructions to correctly set up ONNX Runtime.

---

Getting Started

Install

go get github.com/plandem/silero-go

Download Model or use included one

mkdir -p models
wget -O models/silero_vad.onnx https://github.com/snakers4/silero-vad/files/7603706/silero_vad.onnx

---

Usage Overview

1. Initialize ONNX Runtime

import "github.com/plandem/silero-go/onnx"

onnx.Init("/path/to/libonnxruntime.so")
defer onnx.Destroy()

2. Create Model and Detector

import (
  "github.com/plandem/silero-go/vad"
  "time"
)

model, _ := vad.NewModel(16000, "models/silero_vad.onnx")
detector, _ := vad.NewDetector(model, vad.Config{
  SpeechThreshold: 0.5,
  MinSilence:      100 * time.Millisecond,
  SpeechPad:       30 * time.Millisecond,
}, func(start, end vad.SampleOffset) {
  // Called on speech start/end
})
defer detector.Destroy()

3. Detect from Audio Source

You can run detection from:

• Microphone input (via miniaudio.Capture)
• WAV files (any reader implementing io.Reader with raw PCM samples)

detector.ReadFrom(myInputSource)

---

Audio Format Requirements

To ensure compatibility with the Silero model:

• Sample Rate: 16kHz
• Channels: Mono (1)
• Format: 32-bit float PCM

Use a converter if your audio input doesn't match. This library includes minimal utilities for conversion using malgo.

---

License

MIT License © 2025 Andrei Gaivoronskii