Default TORCH_NUM_THREADS=1 so 14-20 workers do not each spawn a 28-thread BLAS pool.
Keep C920 stereo through SpeechBrain GCC-PHAT Delay-and-Sum, then MetricGAN, and publish mono to LiveKit. Record last_dt_ms and log process_ms/capture_ms every 50 hops for the latency plan.
root