IBM to Open Voice Recognition Software

phug writes "According to the NY Times, IBM is donating code that it estimates cost the company $10 million to develop. One collection of speech software for handling basic words for dates, time and locations, like cities and states, will go to the Apache Software Foundation. The company is also contributing speech-editing tools to a second open-source group, the Eclipse Foundation." There's not much information out there yet - e.g. no word on licenses etc. It is worth pointing out that the Eclipse Foundation was started by IBM.

Comments

Great news

[wertarbyte]

This is great, ViaVoice has disappeared for quite a while now on linux, I hope that this will open a great variety of cool open source applications. If this will be made modular like e.g. festival, I can think of endless applications worth using it.

Re:Great news

[sgant]

This IS great news because I've been trying to talk into my mouse now for quite a while.

"Computer?.....commmm-PU-terrrrr?"

Now hopefully my co-workers will stop giving me strange looks...well, one can dream can't they? No, I'm asking...can one dream?

Re:Great news

[Gentlewhisper]

I think we will see a lot of cool applications for this like virtual ticket sales counters/telemarketing calls (ask a question through the phone and the computer will look up an answer) as well as tech support phone centres!

No need to outsource to India, opensource it to Linux & ViaVoice!

Woohoo! +1 for IBM again!

Re:Great news

[zeromemory]

The last thing I would like to see is a real person replaced with a voice-recognizing computer.

The Alameda County (AC) Transit information number here in the Bay Area uses a voice recognition software to address customer inquiries. The system is very buggy and impractical:

1. Voice recognition is far from perfect. Try getting a computer to recognize the name of a destination or complicated query while you yell it over ambient noise (ex. traffic noise around a bus stop) on a cell phone.

2. The software can only recognize pre-programmed phrases and words. If a user says something that the unexpected, they'll probably get nowhere. Sure, you could present the caller with a list of options, but then why not just make them dial-in options?

Just to compound the problem, the AC Transit system doesn't even provide the option to talk to a real-person!

ViaVoice

[cerberusss]

Is this ViaVoice? The linux packages have been pulled off the IBM site a year or so ago but they're still floating around.

Re:ViaVoice

[sibtrag]

Not likely.

ViaVoice is a wide-vocabulary speech recognition. The article hints at more focused set of target words (times, dates, locations) for the donated package. Sounds much more like the software supporting airlines which use voice recognition systems to help you request flight information.

The strategies are quite different.

ViaVoice encourages you invest some of your time reading training scripts so it can learn your voice and thus recognize a wide variety of words from your specific voice.

The time/date/city system is likely to be speaker independent (no training scripts to read) but much smaller vocabulary.

Re:ViaVoice

[clone22]

I think you're correct. In this type of voice recognition you define a grammar that establishes the words the application expects to hear in a particular state. A state transition occurs when a response matching the grammar is heard.

Obligatory quote

[ssssmemyself]

Are you sure you meant to say "All your base are belong to us?" Did you mean "All you lasers are better than us?"

Code-by-voice

[Max Romantschuk]

Eclipse is actually a kind-of Swiss Army Chainsaw -IDE. You can make plugins for pretty much everything, so one could speculate that a voice recognition plugin would be feasible.

I don't know about everyone else, but the concept of coding by voice does fascinate me. There are obvious issues (like eliminating having to say every single control character (if at all possible)), but with a background of RSI I think it's at least worth a shot.

Thoughts?

Re:Code-by-voice

[Max Romantschuk]

One more thing I forgot to mention in the parent:

Given the fact that most languages have a rather limited vocabulary, and the fact that class libraries and defined functions/variables can be extracted from existing code software like this could make educated guesses on what you were trying to say.

Re:Code-by-voice

[dJOEK]

Eclipse is known for it's good GUI api (or at least it's better than regular swing)

the only way to make voice commands work is to integrate them into your GUI

so your OK-button object does not only have a textlabel-value but also an audiolabel.

this works both ways, one way for accessibility ('hear' what button you will click) and the other way is using your own voice to 'click' it (by saying 'Ok')

Re:Code-by-voice

[LousyPhreak]

this would be nothing more than a nice wow-effect, because most coders write code much faster than speaking it

Re:Code-by-voice

[noselasd]

Thoughts?

I don't quite see myself sitting at my office going,(read out loud)
if parenthese parenthese invar bitwiseor zero x three parentheseend equals
zero or i less than zero parentheseend curlybrace ...
No btw erase from first parenthese on last line to second parentheseend ... arrgh.. invar has a capital V btw.

At any rate I'd guess I can write code faster than I can talk it.

Re:Code-by-voice

[JohnFluxx]

Insightful?
The parent even mentioned RSI. Not many coders with bad RSI can type faster than speaking it :P

IBM is great!

[charlie763]

I love you, IBM. I want you inside me.

Why?

[Anonymous Coward]

Why is it doing this, is it because they think they can make more money with increased software sales? It also might be an advertising campaign, $10 million donation is buying a lot of free coverage.

Corporations dont usually give a way stuff for nothing, in fact their mission by law is to maximize profit.

Re:Why?

[vidnet]

The software they're releasing is probably a project they've given up on (since they have the much more developed ViaVoice engine). Instead of letting it rot in a closet like most companies would, they give it away and score an immense amount of geek points in the process.

Re:Why?

[toolio]

Because this same plan has succeded in the past. Take Eclipse, IBM donated the framework for improvement by the open-source community and use it to build their prioprietory Websphere package. They don't have to worry about improving the API, just the extra features that make it worthwhile to buy.

The same scenario applies here.

Re:Why?

[Anonymous Coward]

IBM is a "solutions company".

They don't make money on software like other companies. The software they develope is used to provide solutions to other people's problems.

Problems they pay IBM to fix. A large portion of the world is now using Linux for stuff. It's free, it's stable, it's as good as a midrange server OS as anything else out there.

They want to use Linux, IBM wants to get their money. So IBM supports Linux.

Also other aspects is what IBM likes. IBM needed a new OS for everything. They have Mainframes, Unix servers, database servers. S/390, Power series, AS/400, etc etc etc.

For a long time IBM dumped money into propriatory software. Once the platform was antiquated, so was their software, and so the millions of dollars of money they put into their own closed source software is a dead end in just a few years. For all the mainframes, database software, developement software, power series, x86, etc etc etc . All these can be fuffilled by Linux. A open source software OS can provide all the functionality that they NEED.

Of course something like OS/400 is better then Linux at running databases, but IBM has the capabilities of making nearly as good. Also this developement also benifits other platforms they support, that OS/400 won't run on.

Buy using Linux they reduce the duplication of effort. No more OS/400 then AIX then this , then that. All of it can be linux, on nearly all their hardware. They just have to make it work.

That's just one of the reasons. They make money from solutions, not software. People buy IBM to make things work, they don't care HOW or WHY, but they want things to work. With Linux they can get things working, cheaper, and eventually cheaper.

No more dumping billions of lines of code into various bits of software that don't integrate and will be obsolete in 3 years. Linux has the potential, thru it's system design and open-ness and flexiblity to never go obsolete.. It'll just change with the times.

Plus IBM would like to see Linux on the desktop, so they can basicly tell microsoft to fuck themselves when time comes.

With this particular bit of software it ties into their websphere and database efforts. Reseptionists can just talk into the computer, people can just talk into the phone, and the computer understands.

But it's worthless without the database and the infrastructure to back it up. If most of the rest of the infrastructure is open source to their customers, why make this little bit of it closed source? It just doesn't make sense.

Sensationalist headlines like "cost IBM 10 million dollars to produce" is misleading.

IBM doesn't give a flying fuck how much money it cost to make it.

There is a well know thing called "sunk cost". It basicly means that money that is spent, is spent. Your not going to get it back. You don't survive long in business if you don't "get" this concept.

A extreme example:

Say you spent 100,000 dollars on a Windows solution. You have found out now that a Linux solution costing 2000 dollars can do what you want, and better.

Your potential to make money on the new system is very high. Your potential to make money on the old system is very low.

Which is smarter? To dump the old software and go with the new to make lots and lots of money? Or to keep the old software just because "you don't want to waste the 100,000 dollars".

A intellegent person will go with the money making sceme and dump the money pit. A stupid person will be blinded by the sacrifice and stick with the old solution because they can't think clearly.

IBM is all about making money. If they figure they can save money by using Linux vs AIX they will. They do recommend it to some of their existing AIX customers...

Think about it this way:
Linux is cheaper and almost as good. IBM saves money, their customers save money. More saved money by IBM customers means that they are more likely to grow and make even more money.

Re:Why?

[Anonymous Coward]

Nice post, but you forget that IBM is a lot more than one company.

They don't make money on software like other companies. The software they develope is used to provide solutions to other people's problems.

No. IBM makes lots of money off software and patents for software processes. WebSphere Application Server, WebSphere Portal Server, Lotus Notes, and of course DB2 make up over a billion dollars in revenue last I heard. Granted, that's less than 5% of IBM's total revenue but it's still income.

They want to use Linux, IBM wants to get their money. So IBM supports Linux.

For IBM Global Services, yes. For Server Group's blade series, yes. For Software Group, hell no. Where is the Lotus Notes client that runs on anything but Windows?

For a long time IBM dumped money into propriatory software. Once the platform was antiquated, so was their software, and so the millions of dollars of money they put into their own closed source software is a dead end in just a few years. For all the mainframes, database software, developement software, power series, x86, etc etc etc . All these can be fuffilled by Linux. A open source software OS can provide all the functionality that they NEED.

No. z/OS has far more capabilities in the traditional business-oriented mainframe space than Linux at present, and it's stupid for IBM to try to push a Unix-like OS into a tightly-controlled mainframe environment. IBM *is* pushing Linux-on-mainframe as a consolidated web hosting environment, but IBM has no plans to kill z/OS.

No more dumping billions of lines of code into various bits of software that don't integrate and will be obsolete in 3 years. Linux has the potential, thru it's system design and open-ness and flexiblity to never go obsolete.. It'll just change with the times.

Not really. First, *lots* of IBM's software never exits the lab, and much that does dies a nasty death in the market. (See Tivoli for dozens of examples.) Second, IBM is riding the Linux bandwagon simply because *it has to* in order to survive.

Plus IBM would like to see Linux on the desktop, so they can basicly tell microsoft to fuck themselves when time comes.

No they don't. If they did they would port Lotus Notes (IBM's flagship desktop application) to Linux.

Sensationalist headlines like "cost IBM 10 million dollars to produce" is misleading.

IBM doesn't give a flying fuck how much money it cost to make it.

IBM does care, a lot, about how much it costs to build something. Let me tell you an IBM internal secret: Eclipse was meant to take down *MS Visual Studio* back in *2000*. Yes, IBM was hoping that Eclipse would *outsell* VS, and when that obviously couln't happen IBM turned it into a marketing win. And lest we forget history already: it took several months of open-source activity before Eclipse was usable by the masses.

Say you spent 100,000 dollars on a Windows solution. You have found out now that a Linux solution costing 2000 dollars can do what you want, and better...

A intellegent person will go with the money making sceme and dump the money pit. A stupid person will be blinded by the sacrifice and stick with the old solution because they can't think clearly.

An intelligent person will evaluate the total business cost of that solution, and ask themselves if they have enough in-house experience to run the Linux solution with the same apparent reliability as the Windows solution. If you've got some *nix talent in-house, the switch is worth it. If you don't have that talent, then the *one-time* cost of $98,000 is more than offset by the continual cost of a new full-time salary.

Think about this: I could go with a cheapo MS MS SQL setup for my company or a expensive IBM database.

Or you could look at the "free" open-source database and cut both Microsoft and IBM out of the picture.

Because it works 99.99995% of the time, an

That means one ore thing missing in linux gone?

[drmancini]

When you look at GNU/Linux as a complex system and think of the things that users complain about when Linux usability is concerned, GPL'd speech recognition software is definitely one of them.

Hooray for IBM and as Ali said in the Linux ad "don't back down"!!

Viable

[tuxter]

Is voice recognition software really viable? When you take into account the different accents, dialect and slang, is it just a pipe dream? Is it a software or hardware related issue?

Re:Viable

[notthepainter]

Yes and no.

I used to work for MacSpeech, we also did large vocabulary dictation systems like ViaVoice.

Back when I was there it really wasn't viable for most people.

However, not all people can type, this includes both the "Hands Free" market (disabilities) and the "Hands Busy" market. Surprisingly, many people also don't want to type, this includes medical and legal professionals. They have an interesting problem, they often need to generate large amounts of boilerplate text quickly. Doctors, Radiologist, Lawyers are also all pretty smart and they heavily use the macro packages to contstruct documentation systems that suit their needs exactly. As you might also imagine, VARS step in and also make these macros.

Is it for you? Maybe not, but it is for a lot of people.

Around 2 decades late...

[Xpilot]

...if only computers (namely Macs) had this technology back in the 80's our favourite 23rd century engineering hero wouldn't have had so much trouble using one at the plexiglass plant. "Hellooooo computer". Still cracks me up.

Human-Centered Computing!

[Milo Fungus]

My brother (who works for IBM) recently sent me an article on USA Today about the system IBM and Honda have developed for speech-interface with a GPS-enabled navigation computer. Really cool stuff.

For those of you who haven't read it, check out The Unfinished Revolution by Michael Dertouzos. I don't agree with all of his analysis (he was a little lacking in pragmatism on some points), but overall this book was very insightful. This book, along with Weaving the Web by Tim Berners-Lee, caused a big paradigm shift in my thinking about computer technology.

Code or training?

[SWroclawski]

In the late 90s I talked with an IBM representative about releasing the ViaVoice source under a Free Software license and the person I talked to (I don't recall his name) said that they might be willing to release the source code- the code wasn't valuable to them. The value in the ViaVoice is the "thousands of hours of training" that the code uses to determine words and voices.

So my question is- will the code released include training to make it work and or will someone be able to put together the necessary resources to train the system.

Re:Code or training?

[maxwell demon]

Maybe set up sort of a ViaVoice@Home project where every geek can help training the software?
Actually it should be quite easy: The client reads your keyboard and the microphone, and you are supposed to speak loudly whatever you type. The training results are regularly exchanged with the central server.

HTK is already availabale as open source

[virtigex]

From the article, it looks like they are making their network grammar version available, not their dictation grammar version. There are types of continuous speech recognition engines, the simple version that uses a hand-crafted network grammar (which seems to be the version that they are talking about), which can be used to recognize simple utterances such as dates, and one that uses a statictical language model and which can recognize an entire language.

This is not earth-shattering news, since HTK has been available for some years. HTK was owned by a company called Entropic and was released as open source when it was bought by Microsoft. HTK can be found at http://htk.eng.cam.ac.uk/. and can handle network grammars. This lessens the impact of IBM's news.

Nice M$-Comment at the end

[echappement]

Nice title;
Speech code from IBM to become open source

And even better.. the comment from Microsoft, quoted at the end of the article
"IBM has not executed in bringing this technology to a broad market as Microsoft has."

Beside the jokes; The article states as well that Microsoft introduced their Speech Server 2004 last March, and that 100,000 software programmers have downloaded Microsoft's free software developers' kit for building speech applications on its Windows .Net technology. What exactly is the difference in quality and approach between the package from M$ and the one here mentioned from IBM ?

IBM also has a grammar based system.

[perky]

IBM also has (or rather had in 98,99,2000) a grammar based recognition system based on the same engine, but using compiled grammars and naturally a cut down acoustic model dependant on the contents of the grammar. There was also a toolset, supporting compiling grammars from BNF, building speech telephony applications and so forth.

IBM Hursley labs had a name dialler 5 years ago that let you phone the computer, say the name fo the person you wanted to speak with, and get put through. They also had a system that provided weather forecasts based on the name of the city or country you said. I was pleased to name the latter "Global Weather Information System" or GWIS, pronounced Gee-whizz. Both ran on the machine under my desk. Both worked reasonably well, especially given that a lot of the acoustic models for names and places were automagically generated.

Re:HTK is NOT availabale as open source

[bonniot]

I was suspicious about MS releasing anything under an Open Source license, so I checked. From HTK's license:

2.1 The Licensor hereby grants the Licensee a non-exclusive license to a) make copies of the Licensed Software in source and object code form for use within the Licensee's organisation; b) modify copies of the Licensed Software to create derivative works thereof for use within the Licensee's organisation.

2.2 The Licensed Software either in whole or in part can not be distributed or sub-licensed to any third party in any form.

This license is in no way Open Source. Yes, you can play with the source, but you cannot build something useful with it and redistribute under the same license.

Good will and a tax deduction

[hrvatska]

Good will in the geek community, free publicity for something that would have just laid around collecting dust otherwise, and maybe a $10 million tax deduction for donating to a non-profit. Not sure about the tax deduction, but this is a donation to a charitable organization, and you can deduct the value of what you donate to these organizations, such as the value of a used car.

Sphinx

[agentk]

Hmm, this is nice, but I was never impressed by ViaVoice. Sphinx is much better to work with.

Reed

Re:HTK is NOT availabale as open source

[Paul Lamere]

For an open-source speech recognition system with a real open source licence check out the CMU Sphinx Project, a family of speech recognition engines, training tools and associated acoustic and language models. The latest version Sphinx-4 is written in Java and is released under a BSD-style license.

Voice software

[RichardX]

Modern voice dictation software is pretty good I'm using viavoice now to write this and I find bark bark shaddup I find that it bark bark shut up damnit bark bark don't make me come down there I find that bark bark okay that's it I'm coming down there argh crash thud bark bark bark bark bark bark

Beer?

[bsartist]

Does this mean that speech is now free as in beer?

Not what you think - here's something that is

[ibi]

They're not open-sourcing anything resembling ViaVoice to the Eclipse folks. Check out the eclipse voice tools proposal. It's directed at making it easier at creating call-center type voice reco apps - not at making Eclipse a voice-directed IDE.

If you're interested in open-source voice recognition check out OSSRI - an effort to bring together some sort of practical large vocab speech recog to linux. They're just starting up, but the mailing list archives hold a fair amount of discussion about the current state of the open-source SR world. (Which, to sum up, isn't that great :-) As a stop-gap they're hoping to get WINE support for Dragon/Scansoft NaturallySpeaking.

In other news,

[mbstone]

Bill Gates announced today that the source code for Microsoft Bob® and Microsoft Clippy®, valued on Microsoft's books at $175 million, has been donated to the Free Software Foundation, a tax-exempt entity.