Posted: Tue Nov 28, 2006 7:42 pm Post subject: [asterisk-speech-rec] Problem with background noise: TipsonS
Tom,
The parameter end_of_speech_timeout will have to be adjusted for the
long audio after you have spoken. For the vad_noise_floor level you
will want to run experiments on what would work for your noisy
environment. There is not a way to adjust this parameter in the
Tuner-Tester because this is a streaming parameter set at the front end
of the Speech Engine.
-----Original Message-----
From: asterisk-speech-rec-bounces@lists.digium.com
[mailto:asterisk-speech-rec-bounces@lists.digium.com] On Behalf Of Tom
[toms400]
Sent: Monday, November 27, 2006 4:26 PM
To: Use of speech recognition in Asterisk
Subject: Re: [asterisk-speech-rec] Problem with background noise:
TipsonSetting Options in Lumenvox.conf
Randal,
Thanks. That's the parameter I've been toying with. I've got it up at
2000, but I'm still having problems. Even though I've uttered a word in
the grammer, the engine continues to accept input until the timeout is
reached - the confidence score is usually 0 in this case.
The utterance may be correctly decoded but there's also a long string of
UHs, BREATHS, LAUGHS, etc.
By answer audio do you mean 'decoded audio'? I'm using the beta version
of the tuner. Is there any way to vary the vad_noise_floor in the
Tuner?
At what level do you typically set vad_noise_floor?
Thanks
On 11/27/06, Randal Rand <RandalRand@lumenvox.com> wrote:
Quote:
Specifically the parameter you would want to increase is the
vad_noise_floor. There are situations where noise in the background,
loud bursting sounds, can hit a certain frequency causing the engine
to recognize it as speech. If you look at your calls with the Tuner
product you can see the requested audio and the answer audio; the
answer audio is after is has be scrubbed and normalized by our Speech
Engine.
Quote:
Also, be sure that confidence scores are in an acceptable threshold
before you allow the call to progress anything below that threshold
should be confirmed or if too low rejected.
[toms400]
Sent: Monday, November 27, 2006 3:54 PM
To: Discussing of speech recognition in Asterisk
Subject: [asterisk-speech-rec] Problem with background noise: Tips
onSetting Options in Lumenvox.conf
I've been having good results with the Lumenvox speech engine when
there is little to no background noise (i.e. people chatting in the
background, cars driving by when walking outside, etc.). But, when
there is such background noise, the results are bordering on
unacceptable. Ideally, I'd like the engine to not regard this noise
as speech input.
I've attempted to change some of the settings in lumenvox.conf, and
was able to get some improvement over the default which was
incredibily sensitive. But, I'm haven't gotten it to a point where
I'm comfortable.
Quote:
Can anyone recommend settings for lumenvox.conf or at least a
systematic way of setting the various options? I've played a little
with the tuner, but don't understand how I can use it in this case.
Thanks
_______________________________________________
--Bandwidth and Colocation provided by Easynews.com --
You can post new topics in this forum You can reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum