Able to reproduce Meta's quality?

Question

Able to reproduce Meta's quality?

listener17 opened this issue a year ago · comments

Did you try to replicate training of Meta?
Just curious - if it is at all possible to replicate the stuff from the code that was shared by Meta?

I would be very curious to hear your opinions.

Thanks!

Zhikang Niu · Answer 1 · Tue May 02 2023 20:07:56 GMT+0800 (China Standard Time)

At the begining, I am a newcomer in speech so I couldn't explained well. And I update some demo, you can listen to those. When training, I'm not add LM model and use the balancer. But I found the result isn't bad.

Zhikang Niu · Answer 2 · Tue May 02 2023 20:12:12 GMT+0800 (China Standard Time)

the audio is used our checkpoint which trained in LibriTTS 960h and 16epochs.
When training, I also found the vq loss probably not very important because it didn't converge...
Maybe there are some small bug

listener17 · Answer 3 · Tue May 02 2023 20:45:04 GMT+0800 (China Standard Time)

Thanks for the insights!