4 ms·

A great narrator can create an audiobook that transcends the text. A terrible narrator can render an audiobook unlistenable. I suspect the robots will eventuall

by thom 4y ago

A great narrator can create an audiobook that transcends the text. A terrible narrator can render an audiobook unlistenable. I suspect the robots will eventually take over (especially if you can tweak voices to your taste), but until then I'll continue to enjoy the likes of Steven Pacey.

ghaff 4y ago
Text-to-speech seems to have hit a certain plateau where it's perfectly functional (for English at any rate). But it would probably take a lot of incremental work to get to network news anchor much less great audiobook narrator territory. There's probably a great deal more value in improving speech-to-text at this point given that AI transcriptions are better than nothing but still quite poor relative to humans.
- thom 4y ago
  I don't doubt that it'll take a lot of incremental work, but look at the rate of progress everywhere else. In 5-10 years, AI speech - coupled with good text models that can understand sentiment and subtext - will probably blow us away.
  - AlbertCory 4y ago
    I don't believe it, but I suppose anything's possible. The narrator has to be a believable person when he/she reads the dialog. I guess that will be the first barrier to fall, followed by a visual AI "person."
- MintsJohn 4y ago
  It's already possible to tag (parts) of sentences on how tts has to speak them, quiet, excited, etc. Soon it'll be just a matter of what is cheaper, a recording or a tagger/director, and that tagging will be partially automatic. As with many things absolute quality, sadly, won't be the target, good enough to keep people buying the product is. Not sure how I feel about this prospect, I love audiobooks, and for big titles the narration cost is a drop in the bucket, and I like the idea of real narration. But at the same time lots of indie books don't get an audio version, and that tts allows more books the have an audio version is a plus and having each character get their own voice works be nice.
  - thom 4y ago
    I think there will always be a premium for human-sourced media. People are going to love being able to tweak the cast and narrator for their favourite audiobooks, but there will always be a market for your favourite celebrity's reading as well. People will swear they can tell the difference between a human versus a machine rendition, and debates will rage the same way they do with audio- and oenophiles. Either way, we'll see within a decade, I expect. I would argue it's much, much simpler than self-driving cars - the text models of today already understand (or at least can classify) a great deal of subtext and sentiment, and the actual audio rendering has become extremely listenable on short texts.

by thom 4y ago

ghaff 4y ago

Text-to-speech seems to have hit a certain plateau where it's perfectly functional (for English at any rate). But it would probably take a lot of incremental work to get to network news anchor much less great audiobook narrator territory. There's probably a great deal more value in improving speech-to-text at this point given that AI transcriptions are better than nothing but still quite poor relative to humans.

thom 4y ago
I don't doubt that it'll take a lot of incremental work, but look at the rate of progress everywhere else. In 5-10 years, AI speech - coupled with good text models that can understand sentiment and subtext - will probably blow us away.
- AlbertCory 4y ago
  I don't believe it, but I suppose anything's possible. The narrator has to be a believable person when he/she reads the dialog. I guess that will be the first barrier to fall, followed by a visual AI "person."
MintsJohn 4y ago
It's already possible to tag (parts) of sentences on how tts has to speak them, quiet, excited, etc. Soon it'll be just a matter of what is cheaper, a recording or a tagger/director, and that tagging will be partially automatic. As with many things absolute quality, sadly, won't be the target, good enough to keep people buying the product is. Not sure how I feel about this prospect, I love audiobooks, and for big titles the narration cost is a drop in the bucket, and I like the idea of real narration. But at the same time lots of indie books don't get an audio version, and that tts allows more books the have an audio version is a plus and having each character get their own voice works be nice.
- thom 4y ago
  I think there will always be a premium for human-sourced media. People are going to love being able to tweak the cast and narrator for their favourite audiobooks, but there will always be a market for your favourite celebrity's reading as well. People will swear they can tell the difference between a human versus a machine rendition, and debates will rage the same way they do with audio- and oenophiles. Either way, we'll see within a decade, I expect. I would argue it's much, much simpler than self-driving cars - the text models of today already understand (or at least can classify) a great deal of subtext and sentiment, and the actual audio rendering has become extremely listenable on short texts.