Unicode symbols from another language

keithito / tacotron

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

MIT License

2.96k stars 959 forks source link

Unicode symbols from another language #278

Open aminbaig opened 5 years ago

aminbaig commented 5 years ago

I am trying to modify the tacotron model for urdu language. I was modifying the symbols.py file and since urdu is a cursive language, if I remove spaces from between the alphabets, they join together and create garbage. Is it ok to have a spave between each symbol?

alchemi5t commented 5 years ago

@aminbaig you do not need the space between the characters. I understand you have joiners and (matras)? but you can have the single amalgamated string, AFAIK. Devanagari is working out for me.

ghost commented 4 years ago

did it work out? @aminbaig

aminbaig commented 4 years ago

Thanks Sahar, I have picked up the project again.

Yes it and thank you for your response. I am now having issues training the data as I get OOM error (out of memory). Any idea whats the minimum amount of memory on the GPU will work with this? I am currently using a Geforce 750 TI with 2 GB ram.