problem with OOD voices (timbre)

[laurentbenaroya]

Thank you for this great work!
I've tried with various OOD voices and I find that the result of GenerSpeech generation gives a timbre that's a bit far from the original (for instance with CMU-speaker-clb-female voice original, parallel output and non parallel output. As for the rhythm, I imagine that this can be corrected.
Is there any way of improving this? Maybe fine-tuning? I don't see a train function in the main branch.