search

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

Spanish vocalization problems #288

Open javierrodenas opened 5 years ago

javierrodenas commented 5 years ago

Hello!!

I have been testing a model with my voice. The voice is very clear and has the same properties as the example voice in spanish that works

The problem is that after 100000 steps, it does not vocalize and is impossible to understand any word. On the other hand, the evaluation wavs sound really well.

What I need to change? As I said, I have the same parameters as a voice that works.

NOTE: My dataset has around 8 hours of audios

Thank you!!!

There are some aligment outputs:

================= 0 =================== alignment-batch_0_sentence_0 linear-batch_0_sentence_0

================= 1 =================== alignment-batch_1_sentence_0 linear-batch_1_sentence_0

================= 2 =================== alignment-batch_2_sentence_0 linear-batch_2_sentence_0

================= 3 =================== alignment-batch_3_sentence_0 linear-batch_3_sentence_0

================= 4 =================== alignment-batch_4_sentence_0 linear-batch_4_sentence_0