[Freeswitch-users] Ways of Integrating Sphinx...

[mszlazak at aol.com]

 BTW Brian,

Here is something that would make FS's VAD much better. The technique also improved Sphinx-3 performance in low-SNR enviroments and made it run over 40% faster.



http://figment.cse.usf.edu/~sfefilat/data/papers/WeBT5.3.pdf


Mark.

-----Original Message-----
From: Brian West <brian at freeswitch.org>
To: freeswitch-users at lists.freeswitch.org
Sent: Sat, 2 May 2009 7:42 am
Subject: Re: [Freeswitch-users] Ways of Integrating Sphinx...












On May 1, 2009, at 6:03 PM, mszlazak at aol.com wrote:


Hi Moiz,

I've checking out mod_pocketshinx against other asr's on Windows with the same hardware.?
No matter what settings one tries, mod_pocketsphinx is virtually unusable in real world scenarios.?





I have used it and it works fine... I think your expectations are a bit high for it... Complex things like dictation is not what PocketSphinx is for. ?You should try linux cuz I know it works great there.


One can play around with mod_pocketsphinx settings so that it picks voice up well but then there better not be any background noise either from a bad connection or just everyday sounds.?





There is no other ASR out there that doesn't get pissed off at background noise or any noise for that matter... have you called AT&T and Sprint lately? ?My dogs barking in the background really send theirs into fits and they paid tons of money for it. ?


It just way to sensitive and of couse you'll notice this problem most with cell phones.





Same with commercial ASR, Granted the acoustical model for PocketSphinx wasn't done with any files recorded from cellphone from what I can tell. ?You can do adaptation of the acoustical model as per the Sphinx wiki to make it more accurate for your needs.... that takes time and effort but it works.


If you adjust the settings to try blocking out background noise you'll find you don't suceed all that well and then there are problems picking up the callers voice.





Those settings are for telling when the person stopped talking... nothing more.


It looks like some kind of signal pre-processing is required that isn't in place yet but we all know that this is a work-in progress.





I'm not working on it... I run the pizza demo with PS and it works from my polycom rather well I would say it gets some things wrong but it does score them low so you can verify it in your scripts.


I don't know if esl would make any difference. To use FS and an ASR/TTS I just bridge calls to another ASR application for now.?

Mark



 



Brian West

brian at freeswitch.org




-- Meet us at ClueCon! ?http://www.cluecon.com







 

=


 





_______________________________________________
Freeswitch-users mailing list
Freeswitch-users at lists.freeswitch.org
http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users
http://www.freeswitch.org



 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-users/attachments/20090504/94029719/attachment-0002.html