Pure managed C# port of WebRTC's voice activity detection (VAD) algorithm.

No native dependencies - works on any platform that supports .NET 8+.

dotnet add package WebRtcVad.NET

using WebRtcVad.NET;

// Create a VAD instance
using var vad = new WebRtcVad
{
    SampleRate = SampleRate.Rate16kHz,
    FrameLength = FrameLength.Length20ms,
    OperatingMode = OperatingMode.Aggressive
};

// Check if audio frame contains speech
byte[] audioFrame = GetAudioFrame(); // 16-bit PCM, little-endian
bool hasSpeech = vad.HasSpeech(audioFrame);

// Or use Span<short> for better performance
Span<short> samples = GetAudioSamples();
bool hasSpeech = vad.HasSpeech(samples);

• SampleRate.Rate8kHz
• SampleRate.Rate16kHz
• SampleRate.Rate32kHz
• SampleRate.Rate48kHz

• FrameLength.Length10ms
• FrameLength.Length20ms
• FrameLength.Length30ms

Higher aggressiveness = fewer false positives (noise detected as speech) but more false negatives (missed speech).

This library only supports raw 16-bit linear PCM audio. It will not work with WAV files or other container formats directly - you need to extract the raw PCM data first.

This is a bit-exact port of the WebRTC VAD algorithm, which uses a Gaussian Mixture Model (GMM) to classify audio frames as speech or non-speech. The algorithm:

1. Splits audio into 6 frequency bands
2. Calculates energy features for each band
3. Uses GMM to compute speech/noise probabilities
4. Applies adaptive thresholds based on operating mode

MIT License - see LICENSE

The algorithm is ported from WebRTC, which is licensed under the WebRTC License.