Re: Voice recognition software?
I've just downloaded and compiled up the latest CVS version of sphinx-4. I
haven't had much time to play with it but I'm quite surprised how accurate it
can be when the number of possible inputs is limited.
Unfortunately, I don't have a very good microphone so the quality of my speech
wasn't brilliant but even so it was able to pick up numbers and menu entries
with a great deal of accuracy. I found that the threshold for starting speech
detection was far to high and required me to nearly have to shout to make it
listen.
If all that was required was picking things from a list / menu / etc I think
sphinx-4 would be good enough. I don't think it would be good enough for
general purpose speech to text applications (but I don't think any speech to
text application comes even close to good enough yet).
Graham
On Friday 13 January 2006 04:55, Jaime Herazo B. wrote:
> * A. F. Cano (afc@shibaya.lonestar.org) wrote:
> > So, let's start this subject rolling,
> > What's the status of speech recognition for Linux these days?
>
> I'm afraid things are still pretty much in the might-be-great-someday
> phase. Last time i checked free software for speech recognition was
> pretty much inexistent. The sphinx guys have advanced, apparently their
> latest sphinx version is much, much better than old stuff, but still no
> dice, and as with such research software some assembly is required (a
> little bit more that just "some" in fact).
>
> Since there's a debian package for sphinx2 (which i've never been able
> to use as anything other than an unreliable toy), i think that making a
> sphinx3.5 package wouldn't be too farfetched, but then again i'm not the
> mantainer. I'd personally settle for a just-apt-get-it-and-works kinda
> package that would sit there waiting for the mic and return any words
> spoken to stdout, just to be able to do some commandline tricks with it
> or to let it sit on some server to give it commands or to make a debian
> version of Serial Experiments Lain's Navi computers or just to brag :)
> The xvoice team was working on replacing their dependence on viavoice by
> replacing it with sphinx, but i think the project's pretty much stagnant
> now.
>
> IMHO full-blown speech recognition in the free software world isn't too
> close, since it depends on lots of research that's behind closed doors.
> If IBM suddenly decided to GPL their viavoice stuff and release free
> data files (the language models i think it is), we'd have full-blown
> free voice recognition everywhere within a year later after that, and
> about 2-3 years later it'd start to really mature as free software, but
> then again i'm just guessing here. Since they've given lots of support
> to linux it's probably not too farfetched, but we don't know the
> internal politics behind it all.
>
>
>
> ---------------------------------------
> "Progress isn't made by early risers. It's made by lazy men trying to
> find easier ways to do something."
> Robert Heinlein
> -----
> Jaime Herazo Barrios /"\
> jherazo_1999 at yahoo dot com \ / ASCII Ribbon Campaign
> ICQ number: 14721935 X Against HTML Mail,
> Yahoo! id: jherazo_1999 / \ and News Too
Reply to: