[mythtv-users] Speaking to control MythTv

[Scot L. Harris]

On Wed, 2005-10-19 at 15:08, Adam Bodnar wrote:
> So I was reading PVR Wire and saw a story about this software for the
> Mac. http://www.speakables.com/TV.htm
> 
> I thought this would be something cool for Myth Tv and from what just
> doing some quick browsing that there are apps to control linux.
> However, I'm not sure if they just launch applications or can issue
> keystrokes.
> 
> Of course the obvious problem would be having your computer trigger on
> sound from the tv. But there are some ways you could work around this
> by adding a trigger word or just have the application not be activated
> when the tv is playing something.
> 
> I could so see my friends thinking I'm insane if they saw me talking to my tv.
> 
> I'm going to look into this a little more, but if anyone has any
> ideas, let me know.


Many many years ago I used something called Vorecone on an Amiga to do
voice control.  The best I could say is that you could get it to work. 
However if there was any background noise it would have to be
retrained.  Things may have progressed in the past few years but there
were some basic problems with voice recognition that IMHO are hard to
resolve in order to get 100% reliability every time.  Most of those
systems required a training session so it would capture your voice input
for the particular command.  If you turned a fan on in the same room
later on it might require retraining.  This would be even more difficult
in a media room where you have surround sound and music or soundtracks
playing at different levels all the time.

I know some of the voice systems the phone company uses now are pretty
good.  They used a huge sample of different people to generate the
matrix used to recognize your voice input.  And in my experience even
those systems have trouble some times.  I personally prefer mashing the
buttons to respond to those systems.

One place I worked tested some voice recognition systems that were used
for access control.  Had the same problems with those due to ambient
background noise.  They used a rolling set of three samples to track any
changes in a persons voice.  Used in the right environment where the
background noises stayed the same at all times it would work reasonably
well.  With any kind of back ground noise it would have problems.  And
if you turned the sensitivity down to low anyone could get in.  :)