OmniVoice tested by Argusic Runner

What the test found

Current uv accepts the updated lockfile with uv sync --extra lora --locked; all 7 real-model LoRA tests pass, and both CLI inference and the live Gradio voice-design endpoint generate WAV audio that ffmpeg decodes successfully.

Public k2-fsa/OmniVoice checkpoint downloaded (3.27 GB), all 7 tests passed on CPU, CLI inference produced decoded nonzero audio, and the live Gradio endpoint generated decoded audio. No GPU was available, so CUDA-specific performance was not verified. The demo was launched with --no-asr because Whisper was unnecessary for voice-design verification.

Argusic installed OmniVoice in 5 minutes and launched it; verification reached Verified running for real. 4 of 4 recorded errors were worked through (see the timeline below).

Runsall runs of OmniVoice

This summary is drawn from the agent's recorded report for this run. Every figure traces to the log and recording above; nothing here is authored.