Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Added "ThaiG2p" & "DIFFS TH" DiffSinger Thai Phonemizer #1274

Open
wants to merge 5 commits into
base: master
Choose a base branch
from

Conversation

spicytigermeat
Copy link
Contributor

@spicytigermeat spicytigermeat commented Sep 7, 2024

  • Trained ThaiG2p on Thai Symbols, WER 0.0396 (99.96% of ~31K words were correctly guessed upon testing the G2p model, this G2p is very accurate! Any errors were only off by 1 or 2 phonemes maximum.)
  • This G2p does use slightly different hyper-parameters and a very different optimizer for training, but in my experiments the results obtained are dramatically more accurate than the default parameters. The G2p.onnx model file is roughly ~14mb. I have not noticed this to cause a slowdown in OpenUTAU at all.
  • Added the G2p to the Lyric Helper & Phonetic Assistant.
  • Included a DiffSinger G2p Phonemizer for Thai. It reads "dsdict-th.yaml" and directly supports Thai characters!

Phonemes used for this G2p and Phonemizer are PrintMov's Phoneme system, which you can see here. There is a public dataset available for training on this dataset, so there is precedent for it to be included in OpenUTAU.

See example using Printto Magicbeat's Thai DiffSinger DB
image

@stakira
Copy link
Owner

stakira commented Oct 27, 2024

Please revert the 30+ files that do not have real changes.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants