search

auspicious3000 / AutoPST

Global Rhythm Style Transfer Without Text Transcriptions
MIT License
260 stars 35 forks source link

the speech content of converted voice with my own trained model changed #14

Open wang1612 opened 2 years ago

wang1612 commented 2 years ago

Why did the speech content of the converted voice with my own trained model changed? Do you know the reason?

auspicious3000 commented 2 years ago

It might be due to the content loss in the content embedding. Maybe you could try replacing the SEA with a better self-supervised model.

wang1612 commented 2 years ago

@auspicious3000 But I did not use SEA and the same problem occurred. Do you know the reason？