Text-to-Speech Synthesis

[Tomato123]

This is a little area of AI that I've always been interested in the progress of. Seen it come from the early days of tacotron to things like 15.ai and Uberduck.

A few free websites:
https://15.ai/ (Probably the best in terms of quality. More voices should be coming soon.)
https://uberduck.ai/ (Not bad quality in some voices, such as Juice WLRD, but the quantity of voices means training was rushed on some.)
https://vo.codes/ (Extremely varying quality. Usually bad.)


I also have attempted to use the "Real-Time-Voice-Cloning" github project by CorentinJ who went on to make the website resemble.ai
While I don't have any examples as this was a while ago, I did get results that somewhat resembled the voice of the G-Man from Half Life when I trained on his voice.
https://github.com/CorentinJ/Real-Time-Voice-Cloning/

 

[Deleted member 568503]

Interesting.

 

[FAST6191]

Yeah I want to get back into this properly.

Found a text to speech program on an amiga many years ago when playing around in workbench rather than launching games. Fascinated me even if it was pretty awful (though I did start making tweaks to words to make it sound like I wanted) and kicked off an interest that lasts to this day.
Keep sort of abreast of it and music generation, but have not got into the weeds. Am hoping someone decides to kick enough supercomputer resources at it to take longer sample lengths as that seems to be the main limiting factor for AI at this point but sample length does not scale linearly in resource requirements.
Might even be worth going back from the linguistics approach again, even if AI is doing all sorts of things.

If we are doing links then somewhat related
https://blog.dshr.org/2018/04/emulating-stephen-hawkings-voice.html
edit might as well have the main article
https://www.sfchronicle.com/bayarea...Valley-quest-to-preserve-Stephen-12759775.php
Well worth reading though