Voice actor here. I've been reading HN since '13. Mostly to stay hip to what's going on in tech as I find it interesting. More recently I've been fascinated by the AI developments in voice / TTS, voice modeling and deep fake video tech.
I'm a software engineer but I work with a lot of projects that require recorded voice narration etc. I've already been able to replace some small/low budget projects with synthesised speech but it definitely does lack the depth and nuance that a real actor brings to it. Is there anything you have seen specifically that you were inpressed by? The one that really stood out to me is the 'deep fake' voice synthesis (though it relies on hours of captured performance first)
Not yet, as you identify it really is about the nuance, and why people like me get paid for what we do. There are some systems that are trying to add generic emotional states using a matrix of activation vs deactivation (how energetic the delivery is) and pleasantness, to try and generate that nuance. So you can end up with emotional states like "guilt"(unpleasant with moderate deactivation) and "elation" ( pleasant with high activation), but as far as I know the application of those states are not determined through machine learning, there is still a human involvement to determine the emotional state, and then try and generate an inflection that matches it.
Narration is as much about acting as having a nice voice. We've only just begun in getting the computer voice to sound mostly human, getting it to act is another thing entirely.
The TTS raises the barrier to entry for new voice actors. Low-paying jobs that they would normally use to "cut their teeth" are being narrated even more poorly by things like Descript and it's ilk.
You'll hear these particularly bad TTS narrations all the time if you suffer through enough Youtube Pre-roll ads.
