Hacker News Reader

Comment by wewewedxfgdf | Hacker News Reader

I'm not convinced TTS can get all the way to the quality of professional actors for things like audiobooks.

I'll take a professional actor over TTS any day - incomparably better quality even with the best TTS.

The technology is still extremely young and immature. As models get better, it should be possible to build tools that allow manual annotations to tweak how much emotion and expressiveness goes into what's being communicated, and eventually it should be possible to fully automate a first pass at this which produces passable results.

In any case, I think the biggest win is that tons of books which have never received audiobooks now have the option of getting a way better alternative than legacy TTS tools. Even if current TTS tools are a bit limited, they still feel like a massive leap in quality from what was available a few years back. Making it trivial to generate better audiobooks will help make tons of information more accessible to people.

The choice of audiobook is rarely going to be between a professional actor and TTS, but between no audiobook at all or a TTS version.

Slightly off-topic, but here’s a video comparing a real voice actor to a mod in a video game. Personally, I think the mod sounds much better.

https://www.youtube.com/watch?v=Ug4h-3qTd1E

Meh - even though the original got negative reactions (for not hitting the mark as a sultry femme fatale), I still think her VO did better readings than a lot of the ones in this. Some of the mod's sound like several different takes by different people spliced together.

(This is still a crazy impressive amount of work, they clearly labored over matching things to facial expressions)

Reply on news.ycombinator.com