[Freeswitch-users] Robust Affordable Speech Recognition

Thu May 13 12:40:31 PDT 2010

Sadly that rules me out as I can't do much with US English. 

Jan

  _____  

From: freeswitch-users-bounces at lists.freeswitch.org
[mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf Of Kashif
Kahn
Sent: 13. mai 2010 20:22
To: freeswitch-users at lists.freeswitch.org
Subject: Re: [Freeswitch-users] Robust Affordable Speech Recognition

Hi Jan,

Our answers to your questions are as follows. Please note that there is a
wide variety of information available on our website under "Speech Engine"
pull-down menu. You should also review Freeswitch section under "Soft-PBX"
for free connector and engine pricing information. Our website is:
http://www.vestec.ca/

1) What languages do you support

We currently support American English and working on major European and
Asian language acoustic models as well.

2) What are your recognition stats per language?

Three points are worthy of note here: 

(a) we have rigorously benchmarked our recognition accuracy against leading
commercial engines in the market and can safely say that we deliver among
the highest recognition accuracy in the industry; 

(b) generally speaking, a native speaker can expect a recognition accuracy -
without "tuning" - in the 90% range while a non-native speaker can expect a
recognition accuracy - again, without "tuning" - in the 80% range. Grammar
"tuning" - for example, via addition of custom pronunciations for difficult
to recognize words - generally improves recognition accuracy by an
additional 5-10%; 

(c) in comparative testing against some of the leading commercial speech
engines, we scored an accuracy improvement of over 3%. See:
http://www.vestec.ca/recog_acc

3) What are the benefits of using this compared to Sphinx/Pocketsphinx?

There are two fundamental advantages of using Vestec over Sphinx: 

(a) Vestec speech engine comes with a robust acoustic model while for Sphinx
you need to develop your own acoustic model. (Acoustic model is the "ear
drum" if you will that does the recognition). Now, development of a high
quality acoustic model is no mean feat, even if you have the necessary
tools. For starters you need high quality training data that can cost tens
of thousands of dollars. Next, you need highly specialized knowledge in
order to properly manipulate the various parameters of training tools for
optimal recognition quality. Finally, you need to be fully prepared to
undertake a trial-and-error process - sometimes spanning 6-8 months - to
achieve the desired recognition accuracy results, with the right data and
training tools. 

(b) Unlike Sphinx, Vestec speech engine supports industry standard grammar
writing format. This allows you to port your existing standard grammars to
the engine as well as make sure that your work is reusable. Similarly, we
provide a variety of useful grammar writing tools and functions that are not
available with Sphinx to make the developer's job a lot easier.

4) How does the (1) installation of a license happen, (2) license check
happen.

You need to put the license files that you purchase from Vestec webstore
under a specific directory. The engine then takes care of the rest. The
license check is host based. In other words, every license is bounded to one
physical/virtual machine. You cannot buy one port license and expect to run
the speech engine on several machines.

5) Does the licensing server support redundancy schemes? 

We used to have a redundancy scheme but recently disabled it in preparation
for the release of a new architecture. The redundancy scheme will be
integrated in the new architecture that will support MRCP. Please note that
software maintenance - covering patches and major upgrades - is free for the
first 12 months since date of license purchase. So, you will be able to
transition to the new architecture over next several months if you were to
purchase a license now.

Hope this helps.

Best regards,
-Kashif

  _____  

From: Jan Berger <jan.berger at video24.no>
To: freeswitch-users at lists.freeswitch.org
Sent: Thu, May 13, 2010 11:36:54 AM
Subject: Re: [Freeswitch-users] Robust Affordable Speech Recognition

Hi,

Hope you don't mind a few nosy questions.

What languages do you support (1) national characters, (2) pre-build
databases.

What are your recognition stats per language?

What are the benefits of using this compared to Sphinx/Pocketsphinx?

How does the (1) installation of a license happen, (2) license check happen.

Does the licensing server support redundancy schemes? 

Jan

  _____  

From: freeswitch-users-bounces at lists.freeswitch.org
[mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf Of Kashif
Kahn
Sent: 13. mai 2010 16:04
To: freeswitch-users at lists.freeswitch.org
Subject: [Freeswitch-users] Robust Affordable Speech Recognition

Dear All,

All those who have wanted a speech recognition solution for Freeswitch but
found the software cost too expensive or the recognition accuracy
unsatisfactory, please consider Vestec Speech Engine for Freeswitch at:
http://www.vestec.ca/products A starter kit - which is a specially priced
one port (ie. one channel) license for the standard engine - is available
for only $25. Additional ports (channels) licenses can be purchased for
$99/port. The engine comes with a free-of-charge Freeswitch connector,
thereby allowing direct interaction via Dialplan.

Best regards,
-Kashif

Kashif Kahn
VP, Business Development
Vestec, Inc.
Waterloo, ON Canada
phone: (519) 885-7615

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-users/attachments/20100513/6026af59/attachment-0001.html 