Best Practice for Voice Conversion with FreeVC

[etlweather]

What are the best practices to get the best voice conversion (converting an existing audio file to sound like a given target audio file)?

1. Is it better to give longer target audio (--target_wav), or shorter?
2. Is there a preferred input audio format? WAV 16bit, 24bit? 48kHz, 44kHz?
3. Should I fine-tune freevc24? How much data and how to format the data for best results?
4. Or should I train from scratch?
5. What is the memory requirements? The freevc24 model isn't big but somehow, this task takes more than 24GB of VRAM on GPU and on CPU it seems to take quite a bit too, like 30GB. Does it need more memory if I have longer audio input?

[eginhard]

1. Try it out
2. FreeVC uses 16 kHz input, anything else will be resampled automatically
3. and 4. The FreeVC training code was never released