search

CorentinJ / Real-Time-Voice-Cloning

Clone a voice in 5 seconds to generate arbitrary speech in real-time
Other
52.05k stars 8.71k forks source link

Italian training #1128

Closed ArdeiG closed 1 year ago

ArdeiG commented 1 year ago

Hi I'm trying to train an italian model for my graduation's thesis. My dataset contains 14gb of multiskeaker's itaian voice

Abaout the encoder: I have done 100k steps and this is the plot result my_runE_umap_106000

about the synthesizer: i have done 33k steps and these are the results

step-33000-mel-spectrogram_sample_1

attention_step_33000_sample_1

about the last picture I'm afraid there are problems because a good output will show a diagonal lines and is not my case, about that I want to know where I may have made mistakes. thanks for the help

auri99 commented 1 year ago

Hello! I've the same problem for the french language ! Any idea?

ArdeiG commented 1 year ago

Hello! I've the same problem for the french language ! Any idea?

I dont know, today I have done more 10k steps, so total is 40k+ steps and this is the result attention_step_42000_sample_1

Alex2610 commented 1 year ago

have you solved the issue? do you have some pre- trained models?