[Asterisk-Users] Anybody using Sphinx

[Chris Albertson]

--- Arnold Ligtvoet <asterisk at ligtvoet.org> wrote:
> Hi,
> 
> I'm trying to get sphinx to work with *. At the moment I believe that
> it
> won't work since there is no audio board in my server and it seems to
> me
> that sphinx expects one.

I've played around with it.  I'm certainly NOT and expert but I'm
pretty shur yu do NOT need a sound card.  Sphinx will be happy to
read an audio recording from a file.

You can train Sphinx on a vocabulary and you can change the
vocabulary database at any time.  So if you ask a "YES/NO"
question you first load the "YES/NO" database.  So if you
train the vocabulary database to only include reasonable
responces you can do well.  One other very powerfull feaure
is that Sphinx can return a set of the most probable matches.
Say you ask the user "when would you like to leave?" and Sphinx
returns (0.50 cow, 0.45 now, 0.05 brow)  Now you can use
external software (your aplication) to scan the list for words
that make sense in the context of your question.  Clearly that
would be "now", not "cow"  For best work you ALWAS want a
propability list and not just it's best guess.

Research shows that humans do just that.  We use context
and grammer more then you'd think.  That's why we tend to
hear what we are expecting and automatically ask people to
repeat if they way something totaly unexpected.

That said yu need good audio quality.  no speech recognition
system is good if there is noise on the line

As for language, Spinx doesn't know or care if you speeak
spanish or Korean it's all just sounds, numbers and symbol
strings.  Spinx does not _understand_ anything.

Spinx would work fine for a dial by name system but you would 
have to plan for the case where it can not find a match or finds
more then one match.  ASking the user to say both first and last
anems may help.  But poor audio quality is the #1 problem and
8Khz sampled "telephone quality" is at the low end of what 
will work..  You'd prefer 16 bit samples and 48Khz sample rate
and a first quality mic.

> 
> Before I continue and try it with an audio board;
> - does sphinx really need an audio interface ?
> - what is the quality like ?
> - will it work in international environments ?
> 
> I was doing a customer installation of Cisco's IPCC express with
> Nuance ASR
> and TTS. The funny thing was that, although the system had some funny
> dutch
> names, the app where you dial an AA and then say the name of the
> person you
> want to speak to works fine for most names.
> 
> I was thinking of creating a simular app for *. Use Festival to
> 'spider'
> through the usernames, do tts, store waves in directory and have
> sphinx
> compare the received names to this. Would this be possible in your
> opinion ?
> 
> TIA,
> Arnold.
> 
> _______________________________________________
> Asterisk-Users mailing list
> Asterisk-Users at lists.digium.com
> http://lists.digium.com/mailman/listinfo/asterisk-users


=====
Chris Albertson
  Home:   310-376-1029  chrisalbertson90278 at yahoo.com
  Cell:   310-990-7550
  Office: 310-336-5189  Christopher.J.Albertson at aero.org
  KG6OMK

__________________________________
Do you Yahoo!?
Protect your identity with Yahoo! Mail AddressGuard
http://antispam.yahoo.com/whatsnewfree