Optimizing and Configuring Speech Generation Models
#2
by
starvano
- opened
Hi! I'm trying to reproduce your result using the RUSLAN dataset in order to further understand how to work with my data.
Could you tell me how many epochs were trained for FastPitch and HiFiGAN? What settings should you pay attention to to get a quality result? Do you do any additional audio processing? Is there any type of communication that is more convenient for you so that you can chat on this topic?
Hi,
@starvano
,
the full training script is here
https://github.com/bene-ges/nemo_compatible/blob/main/scripts/tts/ru_ipa_fastpitch_hifigan/train.sh
Also you can initialize from my checkpoints, it should work.
My telegram account: https://t.me/alexandra_al_a