There is no speaker id in my dataset, what should I do?

[fancat-programer]

No description provided.

[ghost]

Familiarize yourself with the LibriSpeech dataset format. Then reformat your dataset to resemble LibriSpeech. This requires you to separate your audio files by speaker.

Hopefully you are provided enough information to do this. If not, you will need a different dataset.

[fancat-programer]

Unfortunately, no... I will try to teach the encoder to LibriSpeech, LibriTTS, voxforge(1,2) and mozilla common voice(Russian, Belarusian, English)

[fancat-programer]

Should I teach the model in more than two languages? Will it affect the quality?

[ghost]

If the symbols in your languages are mutually exclusive, it should be possible. However, I wouldn't recommend it until you have a lot of experience with training models.

[fancat-programer]

If the symbols in your languages are mutually exclusive, it should be possible. However, I wouldn't recommend it until you have a lot of experience with training models.

By model I mean encoder

[ghost]

It was attempted in #126. Although the encoder model performed very well for speaker verification, the results for voice cloning were not good in that one instance.