f-denkena/livekit-cameras
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: OpenVINO CPU MetricGAN mask; 100 ms hop

Browse Source 
Per-worker OpenVINO CPU runs the MetricGAN BLSTM mask; SpeechBrain
keeps STFT/ISTFT. ENHANCE_INFER=auto with torch fallback. Hop is
0.1 s after isolated last_ms ~6 ms (model+ov+beamform).
This commit is contained in:
root
2026-08-28 07:25:30 +00:00
parent 42443420be
commit ef9573c294
9 changed files with 211 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.env.example
+4 -2
View File
@@ -32,9 +32,11 @@ AUDIO_CHANNELS=1
ENHANCE_MODE=auto
# speechbrain model repo (must be 16 kHz; metricGAN-plus runs ~20-40ms/frame on CPU)
ENHANCE_MODEL=speechbrain/metricgan-plus-voicebank
# model window vs publish hop: 1 s context, 200 ms emit
# model window vs publish hop: 1 s context, 100 ms emit
ENHANCE_CHUNK_S=1.0
ENHANCE_HOP_S=0.2
ENHANCE_HOP_S=0.1
# MetricGAN mask: auto (OpenVINO CPU, Torch fallback) | openvino | torch
ENHANCE_INFER=auto
# C920 stereo Delay-and-Sum (SpeechBrain GCC-PHAT) before MetricGAN
# auto (default): beamform if speechbrain loads, else average channels
# force: require DelaySum_Beamformer, off: average L/R