Synthesizer V to Retrieval Based Voice Conversion

A while back I asked if we could create a voice for SV. I have 20 songs of a singer that has died and I wanted to bring him back to life. Well, Using SV Kevin and then Retrieval Based Voice Conversion I did it! I have to admit, it’s a little creepy. There was another singer we recorded for this song but he had way too much vibrato so SV Kevin took over the job. Thanks SV! :slight_smile:
Crappy song but this is the result and it sounds exactly like the real singer except it’s Kevin’s vibrato.

It takes quite a bit of time to train a voice, especially with 20 songs. I’m not sure what the smallest amount of vocal sample to train it is. That mp3 only ran 110 out of a possible 1,000 training epocs. More epocs improve quality. It takes about 1 hour for 50.

This guy does an excellent tutorial. WARNING: It isn’t easy.

I’m using SV in combo with rvc for a while now (and with so-vits before it).
Leads to quite amazing results.

I tried this as well, over the weekend and was really shocked at the results.