VoxCPM2 /v1/audio/speech returns blank 0.16s WAV contains no sound.
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 37.8k
- Forks
- 4.3k
- Avg merge
- 7m
- Merged PRs (30d)
- 1
Description
vLLM Version: 0.19.0
vLLM-Omni Version: 0.19.0rc2.dev275+ge375b1268
git sha: e375b1268
VoxCPM2 served via vllm serve --omni returns HTTP 200 with content-type: audio/wav,
but the WAV file is only 15,404 bytes (~0.16 seconds) of completely blank/silent audio.
The same model works correctly with the offline end2end.py script (produces valid 3.52s WAV),
but takes 38.6 seconds inference time
Questions:
Is there a known issue with /v1/audio/speech returning blank/silent WAV for VoxCPM2?
Is ref_audio the correct top-level JSON parameter for voice cloning, or should it be passed differently?
Do you have a reference script that loads the model once and benchmarks HTTP inference speed end-to-end?
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Reproduce the request against vllm serve --omni and compare its /v1/audio/speech output with the offline end2end.py result. Check how the ref_audio parameter is passed and measure the HTTP path end to end. Done means the served VoxCPM2 request produces non-silent audio with the expected duration and the parameter behavior and benchmark path are documented.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- api, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 42/100