keithito / tacotron

A TensorFlow implementation of Google's Tacotron speech synthesis with pre-trained model (unofficial)
MIT License
2.96k stars 956 forks source link

output audio lenght #343

Open Ashbajawed opened 4 years ago

Ashbajawed commented 4 years ago

I am training a model on male voice data. training is going on but when I ran demo_server.py it generate audios of 6sec irrespective of lenght if sentence.

e.g for shorter sentences it give audios with silence and for longer sentences it speak out very fast

PS: previously i train it on female voice data (LJSpeech+blizzard) it works perfectly fine on it.. and I also checked my dataset this isnt the case of all audios of dataset of lenght 6 sec in simpler words dataset contain audios of variable lenght from 0-14 secs