Question about the size and training of the FARGAN model
Nobody has claimed this yet.
- Dominant language
- C
- Stars
- 3.3k
- Forks
- 808
- PR merge metrics
- No merged PRs in 30d
Description
Hi～ This issue mentions the model size of FARGAN, which is about 600 MFLOPS complexity:
https://github.com/xiph/LPCNet/issues/215#issue-2580480031
1.
I would like to ask how can I further reduce the model complexity of FARGAN?
For packet loss scenarios, FARGAN only focuses on audio synthesis quality, while compensation quality is mainly handled by the PLCmodel. Is it possible to achieve similar results to the original FARGAN network by training a FARGAN network with fewer convolutions?
2.
Another question is about the adversarial training. I found that the loss didn't decrease and the loss of epoch1 is less than epoch50. Does that make sense?
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
No source file, test, or entry point is identified. Start with the linked LPCNet issue and the FARGAN and PLCmodel training context described here; determine whether reduced convolutions and the reported adversarial-loss behavior are actionable project changes. Done would require a defined scope and reproducible evaluation criteria.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- machine-learning
- Domain
- audio-video-rtc, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100