kantodel

This is a Piper TTS model trained from scratch on youtube transcriptions with cut out silence with ffmpeg, then transcribed with whisper, of Kantorkel.

The model has been currently trained to 10000 epochs on a 3090. The purpose of this model was a fun project at 38c3, a Mastodon bot that you can toot massges at and it will reply with the spoken wav file.

you can find the sourcecode of that project here: https://github.com/nullnullvier/text2torkel

You can find examples or get your own message spoken by tooting at: https://chaos.social/@text2torkel