The best kittens, technology, and video games blog in the world.

Friday, May 07, 2010

Audible review

dia mundial do rock by deadoll from flickr (CC-NC-SA)

Before the posting frequency either reaches singularity or crashes like Nick Clegg's hopes, here's a brief review of Audible.

I have listened to ridiculous amounts of audiobooks (a term I'm going to use for essentially any spoken word - whether an original material or a spoken version of a paper book) for many years now - a very long time before they became anything even vaguely approaching mainstream. And obviously I got most of them from Linux ISO download sites, not counting a very large number which were officially free as podcasts and such.

Then, my cat threw my iPod into a bathtub full of water, I bought a new and much better MP3 player, and with it came free subscription to Audible, which I actually extended for a few more paid months more out of curiosity how this world of paid audiobooks operates than anything else.


First - DRM is as bad as you'd expect. Devices cannot be activate on anything except Windows (the stories about iTunes being able to activate MP3 players are a lie). Then after I formatted the MP3 player, it lost activation and it was impossible to get it back. There was no error message, nothing. Activated successfully, still doesn't work. I had to email Audible for them to fix it... it was really one big pile of ridiculousness, all horror stories about DRM basically came true.

Does it at least DRM the books? Not at all. First, they left CD burning as an option, so anyone who is desperate enough can get virtual CDs and then rip them. And you can download Audible-ripping software which will happily hack into Audible drivers on any activated PC, and convert all your DRMed files into MP3s, FLACs, or whatever. Essentially they fuck over paid customers, and the pirates will get their warez anyway.

This is all good enough reason for not using Audible even if they didn't suck in other ways, but suck they do...

Speech synthesis

But before I get to the main dish of Audible suckiness, a brief interlude for state of speech synthesis. 60 years ago, when computing started, people believed that in no time computers will be as smart as humans. You know - 10 years, 20 tops. Which as it turned out in computing world means essentially "never".

I'm just amazed at how filled with wrong are predictions of people who expect rapid arrival of AI - like most users of the ironically named LessWrong community blog. How the fuck are we going to see AI go "foom" if during the last 60 years there has been essentially zero progress at:
  • Speech recognition
  • Handwriting (and most importantly hand-written math/diagrams/mindmaps) recognition
  • Speech synthesis
Yes, they expect a big foom - but better be worried in case this foomy future Kindle 2020 or whatnot will turn out to be unfriendly - and let's say will read the latest Twilight sequel in a very ironic voice...

The truth is - automated speech synthesis of anywhere near human quality is not coming anytime soon. It just isn't. Amazon knows it as well, while building gimmicky crappy speech synthesis into Kindle, they also purchased Audible for $300M whose business model totally relies on speech synthesis being crappy for decades to come! How is this $300M for a prediction market? Are you shorting Amazon stocks already? I didn't think so.

And while I'm at it, remember how one of my favourite crackpots - Raymond Kurzweil boldly predicted that by 2009 "[people] interact with their computers primarily by voice and by pointing with a device that looks like a pencil. Keyboards still exist but most textual language is created by speaking". Obviously this crackpottery never came true, I'll keep my Microsoft Natural Ergonomic Keyboard 4000 thank you very much. I doubt even Apple would be stupid enough to go into the whole voice thing - but then with iPad they sort of have proven that they can sell absolutely anything, and hipsters will buy it...

iCat by Sontra from flickr (CC-BY)

Why Audible sucks

And now the main part - why Audible sucks. To explain this think for a moment what an Internet shop is. Yes, it needs to have some stuff to sell, and some way to deliver it, and some way to accept payment, but primarily:
An Internet shop is primarily a search engine

That's right! Everything else is just an insignificant attachment to the main functionality of letting people find the stuff they want to buy. Go to Amazon - the archetypal Internet shop. What's there - a huge search engine. With user reviews, recommendations based on your previous purchases, and purchases of other users, and a huge wealth of information which helps you find things you might want to buy. Why nobody has copied Amazon yet? Well, there are some economies of scales involved, but dis-economies of scale exist as well. This is not the answer.

The primary reason is that nobody else has this massive database of data which helps people find what they want. Even if you had all the products Amazon has, at identical prices and terms of delivery, without user reviews and contextual recommendations your shop would be hopeless. And without even as much as Amazon search system? Forget about it.

By the way, I'm probably biased as most of my commercial career was in one way or another related to building custom search engines, just saying for the sake of full disclosure...

Anyway, Audible fails so hard at search it's not even funny. You go to category - and it's sorted by release date. How is this sane? You can sort by customer review - but most of their books have at most one review and there's no Bayesian filtering applied, so everything on the top will be audiobooks with a single 5-star review. Yes, it's all great that ONE person liked it, but it has extremely low predictive power on how good it actually is. About the only useful way to sort anything is by top selling.

So what do we see - title, author, price, star rating, a picture (wtf? these are audiobooks, these pictures are essentially just some random stock photos of no relationship with the product) - and that's it. And max displayed at once is 30 - but don't worry, it will change back to the default useless value of 10 results as soon as you change anything.

This is not Google! You don't want to go from search term out of their site in less than a second - you want to figure out what's actually any good! The logic of having smaller number of results so they're faster doesn't apply at all!

So how do you figure out if the book is any good at all? Well, there are very rarely any customer reviews (just a few star ratings is typical) and the ones there are are typically as useless as "Took me a little while to get into this book, but when I did I absolutely loved it! Have just downloaded the second book and can't wait to begin!". How does this provide any information whatsoever?

There is usually a short publisher-provided blurb telling what the book is about - and you can probably guess what I think about such publishers' honesty.

The only thing that's of any use at all - and without which the entire Audible website would be utterly useless - is the audiobook sample. Unfortunately it doesn't work terribly well. One thing you can figure out is that quality of recording tends to be top notch - unlike some other sources which compress audiobooks as if people were still downloading them on dial-up modems, audible handles quality properly. And you can also find out if you like the narrator or not. Which might be very highly relevant for their "erotic" section at least... (seriously, go there for some major lulz and/or facepalming..., but then maybe I'm not the target audience...).

What you'll rarely find out is if the book is actually any good, which is what we want to establish. Random pages from the middle of the book rarely tell you that. Even worse, I've seen quite a few books for which the random sample was actually legal blurb + acknowledgments from the first few pages... that definitely is some fascinating book, right !?

Unfortunately that's all your search options. Audible won't search inside books to figure out what they're about (speech recognition obviously being crap, but on the other hand - they have access to written original most of the time, so that's a weak excuse), and there's no other information on their website.

If you want to find out a good audiobook, you need to check every audiobook's written equivalent's page on Amazon - for reviews there. Somebody should write a Greasemonkey script for that... (this problem also affects UK Amazon to some extent - often US version has 10x as many reviews, but there's no link and the only way to get to those reviews is Googling... it gets me raging every time...)

How difficult would it be to link it up with the rest of Amazon? Seriously now.

There's also another issue - the choice on Audible is simply less than on "Linux ISOs download sites" - so when you want an audiobook it's very likely you won't find it there, even if it exists. The opposite - Audible having something which pirates don't also happens, but usually for more obscure titles.

This is actually an opportunity for them - sure, many people will torrent Dan Brown, but there's long tail to explore - except it is exactly this long tail which is it direst need of good search functionality. And so Audible fails.

So maybe it's a good idea to short them after all...


Unknown said...

You have a commercial carrier? Neat! What's it do, carry BA B747s?

(I'm terribly sorry, but I just couldn't help it.)

Andy said...

this problem also affects UK Amazon to some extent - often US version has 10x as many reviews, but there's no link and the only way to get to those reviews is Googling

Actually, there is an easier way than that: go to the URL bar, and change the to .com, and you'll end up on the US page for the same item. It seems to work for pretty much anything that exists on both sites.

You can also go the other way, of course, from Amazon US to Amazon UK. If you do it a lot, it's easy enough to create a bookmarklet to do the switch for you.

Quickshot said...

You realize that in your AI research example that you are making the logical fallacy of assuming that past performance will always be a guide to future performance, right?

Going even further off topic, you do realize AI research hasn't actually been stagnant but has been making slow but small gains over the decades, right?

taw said...

Quickshot: It's not logical fallacy, it's called basic forecasting. Past performance is the best predictor of future performance we have.

And I don't believe in AI progress. They haven't done even as little as speech synthesis right, what chance is there for harder problems like automated translation, not to mention actual thinking? The only way serious AI can arrive anytime soon if the progress would look like: 100 years of nothing, everything at once in a few yeas. But such pattern of development is extremely rare in any domain.

Quickshot said...

Ehh, while it might be the best predictor, it's also an utter failure in the face of scientific developments. After all, nuclear fission power plants pretty much happened exactly like you de4scribed there. Not much progress and then a ton within a decade and boom there they were.

Basically for science developments there is no real historical data to base future predictions on. Maybe if we had developed a super advanced civilization before, but if we had, we'd be cheating and cutting all the corners. As such you have no past performance to base future performance on and basically a lot of developments in science will thus seem to come out of nowhere. Well atleast for that method.

More productive would be to look at which bits and pieces they are chipping away at and how much they might already know about intelligence. Or for that matter how capable they are at reaching the resource levels they need.

In that light it's obvious computing power is a lot closer now for instance then it used to be. So the most basic resource is getting there, as for subtasks, visual recognition? A lot better then it used to be, we've reached the level of self navigating cars. These are admittedly not very good yet, but they do work. Voice recognition lately has been a bit stagnant I believe, voice synthesis has improved over time I thought, but I guess not all to much. Specialised creative systems have occasionally lately made appearances. Like auto generation of passable music, even in the styles of old masters.

As such it seems obvious to me that they are continuing to chip away at the limitations, albeit at a some what slow pace. Still, eventually they'll have eliminated all the barriers and 'suddenly' it will be there, with little real warning in past performance metrics.

Which brings me back to the point of saying it's a logical fallacy, you can't use that logic in this particular case.

taw said...

Auto-generated passable music, like this?

History of fission:

1933 Szilárd comes with the idea of nuclear chain reaction

1934 First nuclear fission in lab

1942 First nucear reactor

1951 First nuclear reactor to generate electricity

1954 First nuclear reactor connected to the electric grid.

So two decades from concept to commercialization, the first decade being spent with global production of uranium, plutonium of such being a few grams a year.

This looks nothing like fussion or AI.

Quickshot said...

Technically the idea for atomic power existed decades earlier, they just didn't have any feasible ideas back then. Thanks to Einstein though for instance, they knew there was a lot of power locked up in atoms, if only one knew a way to get it out of them.

And that music generator wasn't the one I meant, I'd heard one made by a music expert that seemingly can make pieces that seem like they come from one of the great composers of past centuries.

As for AI and Fusion, the problem space I guess is just harder then expected, however we have irrevocable proof both are possible. Namely Humans and the Sun, so it is more a question of how. Admittedly the Sun doesn't cut it for commercial power generation, but Fusion has over the decades made clear progress in power gains and if the progress continues at current pace, then probably in a few decades time they'll reach commercially viable returns. But that assumes the past is the best predictor for future events.

taw said...

Quickshot: Humans might be evidence that AI is in principle possible, but they are very weak evidence that we can engineer AI anytime soon.

But even worse - Sun is absolutely no evidence that fusion energy is possible. Sun uses gravitational confinement - something we obviously cannot do - and as far as we know there was never a single case of sustained fusion reaction without gravitational confinement in entire history of the universe.

What's worse - Sun has ridiculously low reaction rate - it takes it literally tens of billions of years to burn through its fuel (otherwise it would be dead already). If we only managed to reach this kinds of reaction rates, fusion would be utterly useless.

Quickshot said...

Yes, I total admit that the Sun fusion rate is utterly lacking. As for self sustaining fusion, ITER is predicted to be able to do that. Even though it's still below the level that we'd consider suitable for commercial power, that would be a huge step forward. Of course they need to prove they can make this next step work as predicted, but all the research they've done up till now supports this line of thinking. Also using your own logic that past performance is the best predictor of the future, this would continue the line of steadily increasing power returns from experimental fusion reactors. As in, this is what you'd conservatively guess from past developments to be the next step.

As for the AI, how sure can you even be that you can actually determine the relevant development branches for it? If you don't know what is actually required you can't really estimate how far off it might be at current rates after all.

Anonymous said...

a WARNING! auidible is *NOT* compatable with many leading products, including SONY !!
- it will NOT work.
I phoned, they basically said 'TOUGH LUCK'!
so CHECK FIRST that ALL your 'equipment' is 'compatable' with audible BEFORE you sign up.
the formats can NOT be changed, so you could be stuck with 'useless' downloads.

Anonymous said...

Warning: Audible is very well known for committing credit card scam/fraud. Search in Google for more details.

Anonymous said...

I agree Audible software is awful .I bought a compatible device which failed to play the files and then the software kept telling me that the device was not connected ,so no chance of trying to reload the files .Audible Said I should format the device ,but the software still fails .I can listen to my books as they appeared on my kindle which is not so wonder pirates flourish .