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

Any ideas for faster testing for audio characteristics (testing without training) #214

Open yoosif0 opened 6 years ago

yoosif0 commented 6 years ago

I wonder if there could be a way where we could test audio features without training training that much (even if the results would be rough). For example, we are prioritizing loss for frequencies less than 4000 Hz. I would like to replace 4000Hz with 6000Hz to check if the quality will improve. Another example is changing the frame_shift_ms in hparams.