I noticed that the model uses Hifigan as the vocoder. How was the hifigan model trained? On what data?
Β· Sign up or log in to comment