[hts-users:04598] Preparing training speech database for Punjabi

[navdeep kaur <nav_783@xxxxxxxxx>]

hi

I am working on building TTS (DNN) for Punajbi Language( Indian). I have to prepared speech corpus for it. I have 8 recording files in .wav format that cover all phonemes.

The punjabi language has 922 phonemes. I need to extract spectral and excitation parameters from them. I have some doubt here. kindly tell whether:

1) first crop all phonemes from available recordings, prepare 922 different .wav files for each extracted phoneme and then determine the parameters values or

2) combine all the .wav files, i have, and then determine parameters after every fixed time interval.

Also tell is there any need to label the database at phoneme level. Which tool i can use to label?

Regards

Navdeep Kaur