onnx-community/silero-vad

Voice Activity Detection • Updated Dec 15, 2024 • 47

Note ONNX export of Silero VAD. Tiny and CPU-only — the usual pick for in-browser and on-device end-pointing.

FireRedTeam/FireRedVAD

Voice Activity Detection • Updated Mar 13 • 4.72k • 65

Note Streaming + non-streaming VAD with audio event detection (speech/singing/music, 100+ languages). Reports 97.57% F1 on FLEURS-VAD-102.

pyannote/segmentation-3.0

Voice Activity Detection • Updated May 10, 2024 • 5.86M • 1.7k

Note Segmentation model behind pyannote.audio 3.x: VAD, overlapped speech, and speaker segmentation from one 10s-window model. Gated.

pyannote/voice-activity-detection

Automatic Speech Recognition • Updated May 10, 2024 • 2.87M • 241

Note Ready-made VAD pipeline over pyannote segmentation — the drop-in option when you want speech regions, not a raw model. Gated.

TEN-framework/ten-vad

Updated Mar 13 • 285 • 146

Note Frame-level VAD for conversational agents; ONNX + WASM/JS builds, and integrated into sherpa-onnx.

funasr/fsmn-vad

Voice Activity Detection • Updated Jun 19 • 3.67k • 24

Note FSMN-based VAD from FunASR — the segmenter in the Paraformer ASR pipelines, strong on Mandarin.

nvidia/Frame_VAD_Multilingual_MarbleNet_v2.0

Voice Activity Detection • Updated May 14, 2025 • 2.74k • 47

Note NeMo frame-level MarbleNet, multilingual. Small CNN, designed for noisy far-field audio.

speechbrain/vad-crdnn-libriparty

Updated Feb 25, 2024 • 13.7k • 44

Note CRDNN VAD trained on LibriParty. The SpeechBrain reference implementation — useful as a research baseline.