How to make this model get transfer-train for other character set?

keithito / tacotron

A TensorFlow implementation of Google's Tacotron speech synthesis with pre-trained model (unofficial)

MIT License

2.96k stars 957 forks source link

Closed hyun06000 closed 6 years ago

hyun06000 commented 6 years ago

First i thank for so wonderful code.

Actually i'm Korean, so i wanna train this model on 'Hangul', the character of our language.

But there is not good datasets. then i made some but it has not enough quantity i think.

That is way i wanna transfer-train, from english to korean.

but when i trained with pretrained step by english for hangul, it gave some error messege about dimension reshaping.

so can i get some hint to transfer?

keithito commented 6 years ago

You can try setting cleaners=transliteration_cleaners in the --hparams flag. It should support transliterating Hangul to ascii, for example:

cleaners.transliteration_cleaners('안녕하세요') 'annyeonghaseyo'