Audiobooks & AI

[DNSB]

I got a chuckle today. I was looking at a link from an email and ran into this gem: How AI is Changing the Way We Read and Discover Books

Traditional audiobook production is expensive and time-consuming, requiring professional voice actors and recording studios. However, AI-driven text-to-speech (TTS) technology has advanced significantly, allowing for natural-sounding, human-like narration.

From the comments I've read on quite a few sites, most people seem to feel that AI generated audiobooks don't come close to even a mediocre voice actor.

I did have to like one of the comments:

It appears that this article was generated by AI...

[Catlady]

Audible is selling audiobooks narrated by "virtual voice." Most of them seem to be in the Plus catalog and/or quite inexpensive, and I suspect most of them are from self-published authors. It's a lot more clutter to wade through.

[rcentros]

Originally Posted by DNSB

From the comments I've read on quite a few sites, most people seem to feel that AI generated audiobooks don't come close to even a mediocre voice actor.

I can always tell when I'm listening to an AI voice on YouTube because they always mispronounce obvious words (usually screwing up long and short vowels in names of famous people). It's kind grating.

[rantanplan]

I think one has to differentiate 2 things here. Does anyone think that AI generated "audiobooks" can replace professionally narrated audiobooks? Not at the moment and probably not for quite some time.

Can AI generated "audiobooks" be used as a tool for people with visual impairments who couldn't otherwise access books that don't have a professional audiobook available? Probably yes.

[DNSB]

As long as it is clearly stated that the audio book was generated using AI/synthesized voices. So far, that does not seem to be the case as in one audio book one acquaintance recently purchased which has all the hallmarks of an AI generated book and lacks any credits for the voicing. Audible seems to think it is worth as much as a human voiced audio book.

[rcentros]

Originally Posted by rantanplan

I think one has to differentiate 2 things here. Does anyone think that AI generated "audiobooks" can replace professionally narrated audiobooks? Not at the moment and probably not for quite some time.

Can AI generated "audiobooks" be used as a tool for people with visual impairments who couldn't otherwise access books that don't have a professional audiobook available? Probably yes.

There are already eReaders that use TTS (Text to Speech). My PocketBook Touch HD 3 will read any ePub to me (with several different voices in English).

[rcentros]

Originally Posted by DNSB

As long as it is clearly stated that the audio book was generated using AI/synthesized voices. So far, that does not seem to be the case as in one audio book one acquaintance recently purchased which has all the hallmarks of an AI generated book and lacks any credits for the voicing. Audible seems to think it is worth as much as a human voiced audio book.

Agree with you here.

[Quoth]

Originally Posted by rantanplan

I think one has to differentiate 2 things here. Does anyone think that AI generated "audiobooks" can replace professionally narrated audiobooks? Not at the moment and probably not for quite some time.

Can AI generated "audiobooks" be used as a tool for people with visual impairments who couldn't otherwise access books that don't have a professional audiobook available? Probably yes.

I tested Google's so called AI audiobook generation. Hardly better than Kindle DXG or Kindle 3 TTS. Oddly no better at all than "local" TTS using Google engine in Android with Pocketbook. Both inferior to 3rd party TTS engine on an old Huawei Android.

That was 2010, nearly 15 years ago. I had TTS on DOS & Windows before 1995 and wasn't much worse.


AI TTS may or may not be be using "AI".


But no, nothing so far comes close. It may not ever because a good narrator undestands the text. AI has no understanding.

[salamanderjuice]

Originally Posted by Quoth

I tested Google's so called AI audiobook generation. Hardly better than Kindle DXG or Kindle 3 TTS. Oddly no better at all than "local" TTS using Google engine in Android with Pocketbook. Both inferior to 3rd party TTS engine on an old Huawei Android.

That was 2010, nearly 15 years ago. I had TTS on DOS & Windows before 1995 and wasn't much worse.


AI TTS may or may not be be using "AI".


But no, nothing so far comes close. It may not ever because a good narrator undestands the text. AI has no understanding.

What 3rd party TTS engine?

The modern stuff is IMO a lot better:

https://suno-ai.notion.site/Bark-Exa...2244ba45ebc2e2

[Quoth]

Originally Posted by salamanderjuice

What 3rd party TTS engine?

The modern stuff is IMO a lot better

The "modern stuff" is not really a lot better than the best 20 to 30 years ago. Sometimes a little better. Degrades quickly if not standard American English.

The good TTS engine was bundled by Hauwei. An SCL-L01 apparently for the Polish market but sold as NOS in Ireland. Pop-in battery, 3.5mm jack socket, SD-Card & SIM slots accessible when battery (cell) popped out. The 2200 mAH cell dated Dec. 2016 (in ISO format). 720 x 1280 pixels. Android 5.1.1

The "flaw" of the mid 1980s TTS I had was it worked best with a custom text file. You could spell phonetically and also add voice modifiers.


I wonder what exactly they mean by AI? Just more real speech sampled? I've an early IC that works with ASCII (and commands) on a simple micro-controller. It's certainly poor, but with a suitable text file not much worse than the DXG or Kindle gen3 Keyboard, or USB audio stuck on a Paperwhite 3. There were better PC TTS on XP 8 years before DXG. XP was decent by 2002. About the same time as Apple Mac OS9 was replaced by much better OSX (based on NeXt Step, based on BSD).

The current Google "AI" effort is only better than PW3 TTS when using standard USA English texts. Picking a different voice doesn't fix non-UAS English.

[salamanderjuice]

Originally Posted by Quoth

The "modern stuff" is not really a lot better than the best 20 to 30 years ago. Sometimes a little better. Degrades quickly if not standard American English.

The good TTS engine was bundled by Hauwei. An SCL-L01 apparently for the Polish market but sold as NOS in Ireland. Pop-in battery, 3.5mm jack socket, SD-Card & SIM slots accessible when battery (cell) popped out. The 2200 mAH cell dated Dec. 2016 (in ISO format). 720 x 1280 pixels. Android 5.1.1

The "flaw" of the mid 1980s TTS I had was it worked best with a custom text file. You could spell phonetically and also add voice modifiers.


I wonder what exactly they mean by AI? Just more real speech sampled? I've an early IC that works with ASCII (and commands) on a simple micro-controller. It's certainly poor, but with a suitable text file not much worse than the DXG or Kindle gen3 Keyboard, or USB audio stuck on a Paperwhite 3. There were better PC TTS on XP 8 years before DXG. XP was decent by 2002. About the same time as Apple Mac OS9 was replaced by much better OSX (based on NeXt Step, based on BSD).

The current Google "AI" effort is only better than PW3 TTS when using standard USA English texts. Picking a different voice doesn't fix non-UAS English.

It's AI because they use neural networks rather than just stringing together phonemes like past approaches (e.g., https://arxiv.org/pdf/1809.08895).

And it's hard disagree from me. The old TTS of 20-30 years ago is awful in comparison. Listen to the Microsoft SAM SAPI5 example on this Wikipedia page: https://en.wikipedia.org/wiki/Micros...-speech_voices. That's what XP was doing. It's not good compared to the more modern Google Android TTS and way way worse than the "AI" approaches.

[Graham44]

Originally Posted by DNSB

As long as it is clearly stated that the audio book was generated using AI/synthesized voices. So far, that does not seem to be the case as in one audio book one acquaintance recently purchased which has all the hallmarks of an AI generated book and lacks any credits for the voicing. Audible seems to think it is worth as much as a human voiced audio book.

The thing is it would have to be clearly stated, not tucked away right down the bottom of the description next to the file size.

i honestly think AI is a complete white elephant, that hopefully will be just quietly dropped over the coming years. You see lots of flashy headlines saying AI can do this, AI can do that, but once you scratch the surface it usually turns out AI can't actually do whatever, or at least not very well. I was reading an article the other week/ month on how they are (or are planning too) using AI in spotting diseases (I think it was Cancer if I remember rightly) in blood samples. Sounds great except when you read the article the results were still being cross checked by humans as the AI made a lot of mistakes - so basically pointless, the medically trained humans might as well have just checked it in the first place.

I know I'm in the very small minority (might just be me on my own ) but I wish AI would disappear, sooner rather than later - rant over*

* Just so there is no confusion, I will qualify my feelings on AI - I hate it with a passion

[Quoth]

Originally Posted by salamanderjuice

And it's hard disagree from me. The old TTS of 20-30 years ago is awful in comparison. Listen to the Microsoft SAM SAPI5 example on this Wikipedia page: https://en.wikipedia.org/wiki/Micros...-speech_voices. That's what XP was doing. It's not good compared to the more modern Google Android TTS and way way worse than the "AI" approaches.

No, that's not what XP was doing. It was what one of the things that MS was doing. I don't need random examples from Wikipedia either. I have a still working XP laptop last re-installed June 2002 and a "clone" of it in a VM on Linux. I actually have a USB audio adaptor and PW3. I actually have a Kindle DXG and Kindle gen3 keyboard.

The advances are marginal. There are several different aspects and so-called Neural Networks / AI don't do much more than the "voice creation". There is zero understanding and the huge overhead of a Neural Net is a brute force approach to issues like lead as in dog or boss and lead as in lining, sinker etc.

*TTS is usable as an accessibility tool, but not as a replacement for a human narrator of Audio books*. At a pinch I can use TTS using Pocketbook on Android. It's certainly the best I've had yet. The offering Google has to "automate" audiobook production for sale is only marginally better for USA standard English texts and no better for anything else. Compare the Hobbit with it and Android phone, Kindle DXG and best XP with a trained human narrator!

Selling audiobooks automatically generated is a kind of fraud. You can do nearly as well on your own phone. It's greed.

[salamanderjuice]

Originally Posted by Quoth

No, that's not what XP was doing. It was what one of the things that MS was doing. I don't need random examples from Wikipedia either. I have a still working XP laptop last re-installed June 2002 and a "clone" of it in a VM on Linux. I actually have a USB audio adaptor and PW3. I actually have a Kindle DXG and Kindle gen3 keyboard.

The advances are marginal. There are several different aspects and so-called Neural Networks / AI don't do much more than the "voice creation". There is zero understanding and the huge overhead of a Neural Net is a brute force approach to issues like lead as in dog or boss and lead as in lining, sinker etc.

*TTS is usable as an accessibility tool, but not as a replacement for a human narrator of Audio books*. At a pinch I can use TTS using Pocketbook on Android. It's certainly the best I've had yet. The offering Google has to "automate" audiobook production for sale is only marginally better for USA standard English texts and no better for anything else. Compare the Hobbit with it and Android phone, Kindle DXG and best XP with a trained human narrator!

Selling audiobooks automatically generated is a kind of fraud. You can do nearly as well on your own phone. It's greed.

Are you talking about some third party TTS that runs on Microsoft Windows XP? If so, which one specifically? Or something built-in to XP? Because Sam was one of the built in voices in Windows XP, they're all similar, and quite similar to the Kindle 2 TTS from what I remember too. And they're all awful and not something I'd want to listen to for an entire book for.

[Quoth]

I'd not listen to an entire book on any TTS. They don't come near human narration. If you are blind or partially sighted they are just about OK for a short story.

Blind & partially sighted have had human narrated audio books since the 19th C. Cassettes are better that CDs as quality is good enough and they remember position.

The rise in audio-books for everyone as digital files is a boon to blind & partially sighted, and some are on our local library website (which is free).

It was a shame that Audible was bought by Amazon and Amazon have hurt the competitors with their aggressive monopolistic marketing. Audible is too expensive and subscriptions are almost a rip-off for most people. So totally unacceptable that Audible is doing computer generated audiobooks. That's only marginally better than say Pocketbook on Android TTS.

I agree that current TTS on a phone is a good bit better than Kindle DXG, kindle gen3 & PW3, but still nowhere near a human reader, especially if content is not standard ordinary USA English. The Kindles mentioned are poor and hardly any improvement on DOS in mid to late 1980s or Windows mid 1990s to today. Win10 is actually more awkward for a blind person than XP and offline speech not enough better. No good to listen to a novel for 4 hours.