search

Plachtaa / seed-vc

State-of-the-Art zero-shot voice conversion & singing voice conversion with in context learning
GNU General Public License v3.0
679 stars 76 forks source link

Training code release dates #5

Open mudabek opened 2 months ago

mudabek commented 2 months ago

Thank you for sharing your work! I wanted to ask when are you planning to release code of seed-vc?

Plachtaa commented 2 months ago

we haven't formulate an optimized training solution yet, so currently there is no exact planned date of training code release

Respaired commented 2 months ago

i'm also very interested in this. I hope the author takes the task of releasing the training code as serious as possible.

liziru commented 2 months ago

Thanks for this great job. Looking forward to the updates.

Selectorrr commented 1 month ago

Unfortunately, I see that none of the authors of popular voice conversion models (DreamVoice, OpenVoice, Diff Heir VC, etc.) publish the code for training. The reasons vary (we are preparing the code, waiting for code acceptance, waiting for review, license issues), but it’s possible they want to retain control over the model for commercial purposes. This is a big problem because all of these models (including Seed VC) share a common issue with accents. It's a localized problem that could be solved through fine-tuning, but the authors don’t address it because it seems insignificant to them. I hope that one day the open-source community will get a truly open model. I hope it will be Seed VC.