Issues training with Cantonese language

keithito / tacotron

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

MIT License

2.94k stars 965 forks source link

Issues training with Cantonese language #306

Open mirfan899 opened 4 years ago

mirfan899 commented 4 years ago

I've updated the symbols.py for training Cantonese. I'm using Jyutping representation of transcripts which English characters with numbers(1-6). Jyutping representation

wong5 si6 faat3 sang1 zoi6 cat1 sap6 ng5 nin4 zi1 cin4 naa5 nin4 maa5 gam1 faa1 sap6 luk6 seoi3

After training the model for 35k steps, eval.py generates 0.4 seconds audio with gibberish sounds.