[Freeswitch-users] Voice to text / speech recognition

Jan Berger jan.berger at video24.no
Mon May 10 15:49:41 PDT 2010


I am going to look into Sphinx myself for IVR menus. 

Recognition factor is not so much the software as it is the quality of the
database and how well it is tuned. Tuning you can do yourself with tools
provided from sphinx + it does exist a separate open source database that I
believe you can use. 

Doing IVR menus is feasible. But, to translate free speech you have a
challenge called dialects. The fact is that recognition factor is not 100%
even with people with the same dialect, and many dialects are very far out
so analyze the language of your clients and make some very accurate
questions to the vendor to ensure you get what you want.


-----Original Message-----
From: freeswitch-users-bounces at lists.freeswitch.org
[mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf Of Wellie
Sent: 10. mai 2010 19:09
To: freeswitch-users at lists.freeswitch.org
Subject: [Freeswitch-users] Voice to text / speech recognition

I'm looking to provide a feature to transcribe voicemails into text for my 
users. I've looked at SpinVox (which was acquired by Nuance). Their 
licensing method doesn't seem particularly friendly or flexible. They 
charge per user and lock each license to a particular user's actual 
identity. If I have a user named John Smith, I have to buy a license from 
Nuance _for John Smith_. If John Smith stops using the voicemail to text 
feature, but I have another user named Jane Doe who wants to start using 
the voicemail to text feature, I have to buy a new license for Jane Doe. 
Since Nuance makes you pay for a license one year at a time, it gets 
expensive if you have any significant churn. Apart from that, it is just 
annoying to have to register users with another company. I'd like to buy 
100 or 1000 or 10000 licenses and use them how I see fit and not have to 
bother with registering individual users with a provider of speech 
recognition services.

I am wondering if other FreeSWITCH users have recommendations for good 
voice to text services or software. I'd be happy to consider either a 
service or software. If software, ideally it would be free of course, but 
a reasonable cost would also be acceptable. I know about PocketSphinx, but 
is it really sufficiently high quality as to be useful for transcribing 
voicemails? I don't need 100% accuracy -- probably 80% or 90% would be 
good enough since I will also attach the WAV file and direct my users to 
use the WAV file when in doubt.

Any pointers or tips would be appreciated.

FreeSWITCH-users mailing list
FreeSWITCH-users at lists.freeswitch.org

More information about the FreeSWITCH-users mailing list