Slashdot Mirror

I had a very similar idea, and it will work. Really. By the way, the poster above, Bysmuth, is dead wrong, labs and all. Feel free to contact me (Bill Cox - waywardgeek@gmail.com) if you need me as a reference to support this idea.

One of my contributions to open source and the blind community has been improving speech speedup algorithms. I listen at > 600 wpm, and have a blind friend who listens at double that. As part of this, I've done numerous A/B tests on many subjects (friends, family and acquaintances), trying to figure out what works for them. Here's what I found. First, anyone who is already a high speed reader also very rapidly becomes able to listen at high speed. This is 100% correlated, after maybe 100-ish tests. I found no counter examples, and the strength of listening speed ability increases with the subject's reading speed. While some speed readers do not hear a voice while reading, it must still using the speech centers in their brain, because high speed readers are already prepared for speed listening, whether they claim to vocalise or not. There are other contributing factors, most notably age. I am the only non-blind person I know who learned to be comfortable speed-listening after the age of 40, though I do have a strong central vision loss issue. Every test I did on with anyone over 40 backed up the fact that speed readers are also naturally speed listeners, but the > 40 crowd is almost violently opposed to speed listening, while the under 40 crowd thinks it's cool. I know... that's such an objective scientific observation :-)

Also, I found that non-blind listeners who force themselves to learn to speed listen (including me), discover that their regular reading speed increases naturally. People can argue all day long about vocalisation being good/bad while reading, but the fact is that the same centers in the brain are used regardless. If you train to listen fast, your reading speed will increase, and vise-versa. This is the single most obvious conclusion I have been able to draw. It's a very real effect.

Another interesting point is that young people will, given a chance, naturally turn up the audio speed over time while listening to good books, very much like we see kids reading faster as they read a good series.

Reading a story both visually and audibly in parallel should enable a reader (whether mostly using their eyes or ears) to focus on the story the way that is more natural for him, and as he goes faster over time, his regular reading speed will increase, regardless of his preference for audio or printed text.

It's slashdot. You're supposed to complain bitterly about the paywall, but when they EDIT it and post the free link, your supposed to lose interest. As evidence, I present every post before this and many follow... which one is actually about the topic of the article? The fixed free link has been posted now for some time.

Since I know these fuckers, I'll tell you who Nuance is. These are the people who buy up companies that get their start helping people with disabilities, and convert them to cash cows that fuck the disabled. They've fucked me three times already, and my ass is pretty sore. In the late 1990's they bought Dragon Systems, who wrote Dragon Dictate, which helped a whole generation of people who couldn't type to use computers by voice. I personally wrote over 1,600 Dragon Dictate/Emacs macros after my RSI injury, and programmed by voice for three years. It was a slow single-word recognition program, and was incapable of either blazing fast text entry or code dictation, but the devs worked very hard to make it work for people who could not type. They were promising continuous command recognition, a feature that would make me likely a faster programmer than with a keyboard, when Nuance bought them. Nuance dumped 100% of their effort on accessibility, and focused solely on the continuous voice dictation into Word market instead. 17 years later, Nuance has not improved voice command and control one iota. This poor guy is just as slow as I was in 1996. Note the long pauses he is forced to make between each command.

My hands mostly recovered, and I type now. However, now I have trouble seeing the computer screen, because I've got some weird central vision problem, similar to Macular Degeneration (but most similar to Stargardt's Disease). So, I learned to listen fast to a computer generated voice which is popular with the blind, called Eloquence. Virtually every highly productive blind computer user in the country uses this text-to-speech engine. I've talked to the primary author, and I know she was highly motivated by how her system could help the blind. She didn't sell directly to Nuance, but after trading hands a couple of times, that's where the #1 voice for the blind is now owned. What does Nuance do with it? They fucking refuse to sell it! It used to be available for ARM processors, and it was awesome on Symbian phones. Now, it's only available bundled with > $1,000 software (JAWs) for the blind, and only in Windows, unless you are OK with using the old IBM version (IBM owned Eloquence for a while) which they generously made available for the blind in Linux. Several attempts have been made by the blind community to enable the blind to continue purchasing Eloquence from Nuance, but they fucking don't care. Now my brain is wired to understand a voice that I can't use. You have no idea how many neurons I devoted to understanding Eloquence at 600 wpm, and my blind friends who can listen twice as fast as me have devoted a significant portion of their brains. The callousness of Nuance is unrivaled.

So... I don't read books in print anymore. Instead I listen to them, spoken by the Mary TTS software, and you have no idea how many hours I spent making that work for me. It's on the order of a man-year, mostly because I couldn't buy Eloquence. I even had to invent a new algorithm to speed up voice by more than 2X with decent quality. I get most of my e-books for almost free from a wonderful service for people with reading disabilities, called Bookshare.org. Unfortunately, my reading interests are not very mainstream, and I often want to read a book available only in print. To do this, I spend $1.50 (yes, only a dollar and a half) at Kinkos (or whatever they are now - I can't read the new sign) to remove the binding but I have to scan the book myself. It's only legal in the US if I do it personally - I'm not allowed to get the scanned version from a blind friend

You are quite right. One thing I've learned over the years is that I should have learned to write well in college, rather than trying to pick it up while being paid to program. I've written SBIR grant proposals (all were funded), various statement of work proposals, and a number of patents. It was damned hard for me, but it had to be one. I'm steering my 10 year old son towards taking writing more seriously, and hopefully he'll not have the same writing handicap I had when entering the workforce.

I will dispute one point. In EDA at least, results are what counts. Once algorithms are written and benchmarks performed showing yours beats other well known algorithms in some area, IMO, you have data worth sharing. This is not how it works today. I put my own hobby research on the web, but as I've said, I gave up on dealing with the PITA journals ages ago. Here's a great algorithm for better speech frequency analysis. Here's a better speech speedup algorithm for > 2X. Promoting algorithms I develop for free is also painful. Getting sonic into Debian was not fun at all, though it seems to have been adopted in Android and several TTS engines almost magically. I believe it's now even in the Android Audible client, which is now far superior than the iOS client for high speed. I can't get either algorithm linked to on Wikipedia because my web pages don't pass their test as a credible source.

How are you supposed to share great ideas?

You are quite right. One thing I've learned over the years is that I should have learned to write well in college, rather than trying to pick it up while being paid to program. I've written SBIR grant proposals (all were funded), various statement of work proposals, and a number of patents. It was damned hard for me, but it had to be one. I'm steering my 10 year old son towards taking writing more seriously, and hopefully he'll not have the same writing handicap I had when entering the workforce.

I will dispute one point. In EDA at least, results are what counts. Once algorithms are written and benchmarks performed showing yours beats other well known algorithms in some area, IMO, you have data worth sharing. This is not how it works today. I put my own hobby research on the web, but as I've said, I gave up on dealing with the PITA journals ages ago. Here's a great algorithm for better speech frequency analysis. Here's a better speech speedup algorithm for > 2X. Promoting algorithms I develop for free is also painful. Getting sonic into Debian was not fun at all, though it seems to have been adopted in Android and several TTS engines almost magically. I believe it's now even in the Android Audible client, which is now far superior than the iOS client for high speed. I can't get either algorithm linked to on Wikipedia because my web pages don't pass their test as a credible source.

How are you supposed to share great ideas?

Well said. I'd say open source projects are successful when they give those contributing what they want. In my case, I volunteer for projects that help blind/VI people, projects like Vinux - Linux for the Vision Impaired, SpeechHub - free voices everywhere, NVDA - the free screen reader for the blind, and Orca - the Linux screen reader. I also contribute algorithms, such as libsonic - speeding up speech for speed listeners, and an enhanced FFT algorithm for speech recognition.

So, my win is helping the blind and otherwise disabled with computing technology. In Open Source Land, it's whatever floats your boat.

http://dev.vinux-project.org./ It has links to my speech speed-up work, sonic, my work with Mary TTS, VInux, and some ideas about improving speech recognition front ends with better speech spectrograms.

Comparing Debian to Android is actually very interesting. Debian has something like 32,000 packages that can be installed, but it's taken something like 15 years to get there. Android blasted to over 100,000 in something like 2 to 3 years. Debian is all about community contribution, while Android is all about selling closed-source apps, with no sharing of code between. In theory, it should be easier to publish an app in Debian than Android, but this is not the case at all. In Debian, you have to find a sponsor, do a complicated job of packaging, pray your package gets uploaded to Unstable, and then wait a few years while it migrates to Stable before other programmers will generally have access to your work. I call this the Debian Red Tape. It's suffocating innovation in the open-source community, and it's the reason Android is kicking Debian butt.

I believe there is a solution, but it requires a completely new packaging system. Let's compare Android and Debian packaging:

- Android ships every dependent binary in the .apk app file. This eliminates the nightmare of having your app crash because some library you use gets updated.
- Debian is all about resusing .so files across applications. This made sense in the '90s when disk space was scarce, but now days, it's just dumb. The reason it takes years to get a packaged library into Debian Stable is that it takes years before we believe you library wont cause other apps to crash.

A new packaging system could share identical binaries between apps to save both disk and memory space, but it should not ever change a binary used by an app. Also, publishing new packages should be as easy as creating a repo on github.net. You simply declare that it's available, and everyone can use it. Whether a developer decides to depend on your code should be a matter of trust, which could be scored based on developer reputation, code stability and what other packages use it.

Without a major upgrade to our packaging system, Debian will continue to fall further and further behind. Why do so many people feel they have to build a custom Debian based distro? Because Debian incapable of addressing the needs of modern users. Frankly, even with the total lack of libraries available for Android, and with Google having their heads up there arse with respect to accepting contributions from the community, I am able to contribute more to Android than I can to Debian. Check out my library that I've made available to both at dev.vinux-project.org/sonic. I'm basically done for Android, while I'm still waiting for a Debian sponsor over in Debian land.