Can modern Text to Speech voices replace professional voice over actors? | Hacker News Reader

Can modern Text to Speech voices replace professional voice over actors?

play.ht

28 pointsh998 comments

I'm always surprised that the computer games industry hasn't latched onto speech synthesis. Modern speech generation coupled with some sort of emotional + accent markup and a library of base voices would allow going from script writer to voice over, without the time and expense of traditional approaches and often poor results. You could even go procedural, dynamically generating speech. The main character can finally get a name and stop being a silent protagonist. Mods & DLC. Include novels' worth of content without the immersion reduction of text boxes.

After recently playing Unavowed I kind of hope there'll always be a place for human acting. That said, for games with no voiceover it would be nice to have an accessible TTS option without monotonous tones.

The answer to the question: no. Not yet. The examples in the article are very impressive, but note that they are also very short. Listen to them more than once. There is a shape to the sentences: they start at a low pitch, rise at the end on clauses and then finally fall at the end of the sentence.

Now imagine those same tones in this text: "Stop. I don't want any of your damn potatoes. If you try coming on to my property again, I will set loose the dogs." It would be perfectly understandable, but comical.

We won't get full text to speech replacement until the computer understands what's in the text and that won't happen until we get GI. We can probably refine it a bit, but it's always going to be a bit odd. With enough text you will always get that uncanny valley effect.

Having said that, probably we can build systems that will allow a single voice actor to play all parts in a production. That could have many advantages and I expect it to happy relatively soon.

I would disagree. Sentiment analysis is already there. Good text to speech too. Basically one needs to use them together to give the speech engine a clue towards the required sentiment and the speech engine enough samples to use different intonations for different sentiments. Won't beat a skilled speech artist, but should be more than sufficient for any mediocre quality cartoon. The big upside is instant translation to all the languages supported by the speech engine. Make mediocre cartoons, add speech, upload to YouTube for ten+ different locales, profit.

> Can modern Text to Speech voices replace professional voice over actors?

It's a bit like saying can super intelligent AI create better Sonic the Hedgehog games.

Once you get nice Text To Speech the fabric of society will change dramatically.

I guess in a video game you can spend hours choosing and modifying parts of the speech, so a professional could do great work with text to speech that in normal society wouldn't work. But it doesn't seem like what they are talking about.

This seems promising, though I think the price is a bit too high.

I also hope there is a service that can automate syndicate the voice track while making a Youtube video, any existing services?

Btw, what do you think this service compare with the amazon polly?

WaveGlow from NVidia released at the end of 2018 is much more impressive: https://nv-adlr.github.io/WaveGlow.

Are there any state of the art improvements after this one?

Is there a best way to go the other way, that is, from speech voice to text?

I’ve found the transcription tools tools to have difficulty with at least the stock computer voices.

Reply on news.ycombinator.com