Slashdot Mirror
Voice-to-Text Options for Unix?
fingerLess asks: "I recently got pushed over the edge in keyboard
use. I use Linux and wanted to find a good voice-to-text solution I
can use on Linux on my laptop. It seems the IBM ViaVoice I found was
still at 1.0 and there was even some questions if it was still available.
But if it isn't being worked on, is it worth it? Has anyone tried any
voice products running on top of one of the win-virtual machines and
had had any success? My experience with those indicate top much of a
performance hit in the AV department (AV seems not to be a real high
priority with such products aimed at business or 'Office'
productivity). Ideas?" For a while there, it looked like speech
recognition was progressing at a pretty good clip, especially with
Big Blue leading the charge. However I haven't heard of anything
revolutionary happening with this technology for the past 2 years.
Did I miss something, or has voice recognition on the desktop lagged.
Just another instance of big blue's underwhelming support. But seriously, they likely consider it more of a server and mainframe OS, as they don't offer it on any of their desktops preinstalled.
Remember, there were no nuclear weapons before women were allowed to vote.
Try to have a look at Emacspeak. Perhaps that's what you want.
Programming can be fun again. Film at 11.
I just found this cool new thing called google. It is a 'search engine', whatever that is. Any ways, it allows you to type in key words that your interested in, and it spits back some web pages with related information. Try it out http://www.google.com
This isn't the sig you are looking for... Carry on...
There is a large vocabulary recogniton system, CMU Sphynx at http://www.speech.cs.cmu.edu/sphinx/
However, This is probably not exactly what you are looking for as is not (yet?) suitable for Voice Recognition tasks.
The problem with Voice Recognition is that it has always been a toy for most users and very few of those who buy Voice Recognition software do succeed to make a productivity boost. If you are one of them, you are a lucky guy as you have a good, distinguishable pronunciation, you work in a silent environment and use the mike shipped with the software. Since Unix world has a very practical view on things, I doubt there is many unix people out there that think Voice Recognition can be of use to them.
Given the laziness of users and lack of training facilities, Voice Recognition is considered to stay an unprofitable buisiness for a long time. You can't even imagine how expensive it is to write a Voice Recognition software and collect the speech data for it...
The title said UNIX, and then later he said he was using Linux, so this may not be as applicable as he wanted.
This morning I saw a review on IBM ViaVoice for MacOS X that piqued my interest. Overall, it looks like a pretty solid product for doing voice input into any program.. but can you imagine using vi without a keyboard?
As a recent MacOS X convert -- it's good to have a UNIX with supported commercial apps.
In my view, one the primary obstacles that has yet to be overcome in the wonderful world of voice recognition (regardless of OS) is the specialized vocabulary that is required by the recognition software. By this, I am specifically referring the word syntax that the interface requires to achieve optimum performance.
While we have all seen the world quite capably adapt to the Palm-Graffiti style of hand writing recognition, many vendors have found it to be a much more formidable task to modify the manner in which people speak. Beyond the several language variations (languages, accents, lisps, etc...), developers must also take into consideration much more subtle disparities in speech such as separate dialects within a given language. This has caused quite an immense dilemma, one that has prevented the mainstreaming of such technology!!!
Even in the case of software such as Via Voice, the user is still given to the quite arduous task of creating a "dictionary" of sorts that recognizes their specific speech patterns and verbiage tendencies.
All of these factors lead to complications and idiosyncrasies that the average Joe User is unwilling or unable to accept!!!
Beer is proof that God loves us and wants us to be happy. -- Benjamin Franklin
This stuff is pretty complex, and I doubt you will find anything that good for free...
"Karma can only be portioned out by the cosmos." -Homer Simpson
Voice recognition hit a 'okay, nothing great' wall approximately three years ago.
Dragon Naturally-Speaking is the Win state-of-the-art. It converts speech to text and you can create macros and you can also speech-activate the keyboard buttons. That does not necessarily means it is right for you. It is good, requires training, requires a BEEFY CPU and RAM, and 'should' get better over time. The problem is that it is slow if you are a fast speaker. This becomes distracting when you are 100 words along and it starts converting words 1-10. And, the dictionaries needed serious additions. I've used versions three through five. Five is good, but it could be much better. Dragon Systems was purchased and rehashed about two years ago. [BEWARE Dragon-authorized service providers who charge extra for what comes in the box!]
IBM ViaVoice-Win is not bad. There were two versions: one trainable, the other stand-alone. The stand-alone was crap for me. The trainable version wasn't too bad. [I have NOT used ViaVoice-Mac.]
Hark is the best of the best
All in all, Voice Recognition is reasonably good but it is not where it should be, yet.
iDictate for Mac OS X (MacSpeech, Inc.)
ViaVoice for Mac OS X (IBM)
I seem to remember that commercial versions of Mandrake up to 8.0 had IBM ViaVoice... I haven't bought a full commercial edition of 8.1, so I don't know if they still do that, though.
ISIP has a pretty good speech to text system that should work on most Linux/Unix boxes.
Takes a little intelligence to set up though.