from Hacker News

Ask HN: How would you build a text to speech startup?

by h3ctic on 7/25/20, 7:43 PM with 1 comments

Since Google's WaveNet TTS systems shouldn't sound robotic but they still do. Third party TTS engines for Android aren't good and those who did sound good have been bought and shut down. Which markets would you approach? What problem of yours can be solved?

by PaulHoule on 7/25/20, 11:04 PM
Imagine you were making a visual novel:
https://en.wikipedia.org/wiki/Visual_novel
you could write one is not voiced, voiced with some exclamations ("Are you sure that's safe?" gets replaced with "Yikes!") voiced in some places and not voiced in others, or fully voiced.
In 2020 you need to hire and direct a voice actor because TTS is nowhere near good enough. If there was a good enough TTS you wouldn't need to hire the voice actor although you'd still need to direct the TTS.
Two of my favorite speakers are Lester Holt and Tanaka Rie and I'd like to see a computer voice that approaches that kind of person.