How we prepared voice cloning TTS to run on-device

We trained a text-to-speech model with voice cloning that fits into 100 MB (fp16) and runs on-device...