![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
Audiobooks by AI
I just had a look again at the collaboration between Project Gutenberg and Microsoft's AI to record audiobooks. They have produced (as far as I can count) 4,840 audiobooks this way, all available free on the Internet Archive and various other places.
https://marhamilresearch4.blob.core.windows.net/gutenberg-public/Website/index.html
I checked out three that interested me:
▪ "My Man Jeeves" by P G Wodehouse
▪ "The Time Machine" by H G Wells
▪ "The Moon Voyage" by Jules Verne
The results were somewhat surprising.
"My Man Jeeves" was terrible. I love P G Wodehouse's stories. His stories have a wonderful, humorous, sing-song quality. A lot of their humor is expressed in the mannerisms of the characters. This was completely lost on the AI. A pity, but not really unexpected. Even if it had used a British accent instead of the totally wrong USA accent, it still wouldn't have worked. The AI has no idea of how to say, "Ripping! I'll be toddling up, then. Toodle-oo, Bertie, old man. See you later."
"The Time Machine" was jarring, with pauses in inappropriate places, and the wrong tone because the AI obviously didn't understand what it was reading. Again, even if it had used an English accent it still would have failed.
"The Moon Voyage", except for it reading out the contents pages for nearly 3 minutes, fared much better. The reading sounded very natural. I'm not sure why. I would have thought it would have encountered the same problems, but perhaps a more modern translation from French helped.
So, okay, I'm holding this to a very high standard, and it could be said that I'm being a bit of a hypocrite. I have often had my Amiga computer read books and articles to me, and its voice uses very primitive text-to-speech, but I was quite happy with that (especially since I hacked it to give it an Australian accent :) ). Even now I have a very simple speech synthesis program on my Linux computer which is similar to the Amiga's and not even close to the capability of Microsoft's AI. But times move on, and Microsoft has billions of dollars to spend on massive computing systems. I live below the poverty line, and have a cheap, crappy, little computer. Of course I expect more from them.
https://marhamilresearch4.blob.core.windows.net/gutenberg-public/Website/index.html
I checked out three that interested me:
▪ "My Man Jeeves" by P G Wodehouse
▪ "The Time Machine" by H G Wells
▪ "The Moon Voyage" by Jules Verne
The results were somewhat surprising.
"My Man Jeeves" was terrible. I love P G Wodehouse's stories. His stories have a wonderful, humorous, sing-song quality. A lot of their humor is expressed in the mannerisms of the characters. This was completely lost on the AI. A pity, but not really unexpected. Even if it had used a British accent instead of the totally wrong USA accent, it still wouldn't have worked. The AI has no idea of how to say, "Ripping! I'll be toddling up, then. Toodle-oo, Bertie, old man. See you later."
"The Time Machine" was jarring, with pauses in inappropriate places, and the wrong tone because the AI obviously didn't understand what it was reading. Again, even if it had used an English accent it still would have failed.
"The Moon Voyage", except for it reading out the contents pages for nearly 3 minutes, fared much better. The reading sounded very natural. I'm not sure why. I would have thought it would have encountered the same problems, but perhaps a more modern translation from French helped.
So, okay, I'm holding this to a very high standard, and it could be said that I'm being a bit of a hypocrite. I have often had my Amiga computer read books and articles to me, and its voice uses very primitive text-to-speech, but I was quite happy with that (especially since I hacked it to give it an Australian accent :) ). Even now I have a very simple speech synthesis program on my Linux computer which is similar to the Amiga's and not even close to the capability of Microsoft's AI. But times move on, and Microsoft has billions of dollars to spend on massive computing systems. I live below the poverty line, and have a cheap, crappy, little computer. Of course I expect more from them.