Sample:https://vocaroo.com/1nvl8SkJ51VG
Tortoise TTS model to use in ai voice cloning repo with an audio sample. It can generate at low samples and comes out better than the stock model. I think I used 32/160 settings for the sample. 96/200 gives better results but of course you are trading computation for quality. may have to clean up extra noises in between long text, as with any tortoise model.
Works very well with RVC applied on top. Much more stable than bark for something like an essay or audiobook.
Trained at full precision for 200 epochs from about 4 hours of data. Loss of about ~1.18