Training Russian TTS with phoneme

TTS (Text-to-Speech)
#1

Hello, I am trying to make russian model with ‘ru’ phome_language.
But it gives an error. Аt the same time I changed the configuration (config.json)

image
image1794×436 131 KB

vocabalary parameters:
Screenshot_20210119_175616
Screenshot_20210119_1756161107×192 39 KB

I use train_tts.
image
image1132×405 49.5 KB

ERROR : [WARNING] 1 utterances containing language switches on lines 1
[WARNING] extra phones may appear in the “ru” phoneset
[WARNING] language switch flags have been removed (applying “remove-flags” policy)
Would appreciate any help!

#2

I’d not worry about it a lot for the first run. Check it back if the model has pronunciation problems later on.

#3

Hi @erogol!
Thanks, but i have training models Tacataron2(200 000 steps) + MultibendMelgan(500 000 steps).

image
image1408×857 296 KB

image
image1343×876 318 KB

Can you check audio ?
russuian_voice.zip (228.9 КБ)
What do you think about include phonemes?

#4

I didn’t check the audio but the graphs look okay. Do you experience any particular problem?