Skip to content

v1.0.0 — Initial Release

Latest
Latest

Choose a tag to compare

View all tags
@beshkenadze beshkenadze released this 05 Mar 20:16
v1.0.0
bdbafe9
This commit was signed with the committer’s verified signature.
beshkenadze Aleksandr Beshkenadze
SSH Key Fingerprint: vdVdFbJFy3rwRYkyksFDfYe3SvMdMZGFDWvuwPPeG8Y
Verified
Learn about vigilant mode.

FireRedVADKit v1.0.0

Swift package for real-time Voice Activity Detection on Apple Neural Engine.

Features

  • DFSMN Stream-VAD model ported from PyTorch to CoreML (neuralNetwork format)
  • Kaldi-compatible 80-mel log-fbank via Accelerate/vDSP
  • Full pipeline parity with Python reference (max diff 5.96e-08)
  • Streaming API: 100ms audio chunks → speech segments
  • 4-state VAD postprocessor

Benchmarks (Apple Silicon)

Compute Avg Latency RTF
All (ANE) 0.260 ms 0.0032x
CPU Only 0.294 ms 0.0037x

Installation

.package(url: "https://github.com/beshkenadze/FireRedVADKit.git", from: "1.0.0")
Assets 2
Loading