We use ncnn to support multiple platform, especially mobile
What’s included
- Stream VAD (frame-by-frame, packed cache)
- Non-stream VAD (whole sequence)
- Non-stream AED (3-class: speech, singing, music)
More details in NCNN README.md
- Rust ONNX Example — Stream VAD using wavekat-vad with pure Rust Mel filterbank + CMVN preprocessing