NVIDIA / NVIDIA/NeMo-Speech.cpp
Magpie TTS 2602 and 2607 for Japanese language inserts hallucinated sounds on punctuation symbols (periods).
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 117
- Forks
- 29
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 6
Description
There are odd sounds inserted where there should be pauses representing CJK periods. It happens for most, but not all periods. The hallucinated sounds to my ear seem to be either mangled Japanese sounds similar to the sounds nearby or seemingly random English-like sounds.
Few things I noticed:
- CJK commas don't cause hallucinations.
- Replacing CJK period with Western period didn't help.
- Final period doesn't cause an issue.
- If you put the period at the beginning of the text, there's an issue.
For reference:
The builds in the releases do not support tokenisation for Japanese language for Magpie TTS. So I built it from source with Japanese support (in my case Windows with CPU backend).
Contributor guide
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
Start by reproducing Japanese Magpie TTS from a source build with Japanese tokenisation enabled, using CJK periods in positions described in the report and comparing commas, Western periods, and final periods. No source files or tests are named; done means punctuation produces pauses without hallucinated sounds across the reported cases.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- audio-video-rtc
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 48/100