VoIP Mailing List Archives

[pete at privateconnect...]

Good morning all,

I realize this is slightly off the FS topic, but I am wondering if anyone out there has experience with software packages designed for the transcription of voicemails to text. I've used pocketsphinx with FS to handle IVR menus, but now have the task of figuring out how to convert recorded phone conversations (voicemails mostly) to text.

This does not have to be a real-time process, I can store the audio files and process them over time. This would need to be a software (preferable open source) solution. ASPs like VoiceCloud would not work for this application.

Thanks for any help
-pete

[dave at 3c.co.uk]

Good evening Pete,

The only way to do this is, I'm afraid, to use a human. We use Amazon's
Mechanical Turk to good effect.

Cheers --

Dave

[pete at privateconnect...]

I apologize, I should have been more clear. We will be using humans to scan the translated results. But we are looking for a system to perform the "first pass" on the audio to hopefully help the human type less.

Although the question has been raised if it's faster to have a human just transcribe the whole thing, or fix up what the computer spit out. If you have any insights on this, that would be great.

-pete

[darren at aleph-com.net]

We've been using mycaption for a number of months and have been very
happy with the results. I'd suggest giving them a try.

Darren Wiebe
darren@aleph-com.net

Pete Mueller wrote:

[dave at 3c.co.uk]

Hi Pete,

I'm afraid that the answer's still the same: use a human. Here's an
article describing the state of the art:
http://www.theregister.co.uk/2009/08/05/spinvox_demo_day/
- the links to previous stories at the bottom provide good background.

--Dave

[kirk.bateman at gmail.com]

I'm still interested in getting pocketsphinx to attempt speech recognition on an audio file.

To be honest, most of the problem is that at 8Khz (mobile phone call rate), speech detection is NOT very accurate, at 16Khz it IS significantly better.

I'm planning to have a play with the speechtools module and mod_pocketsphinx etc to try and get an audio file parsed, spare time permitting.

Will let the list know if I get anywhere.

Regards

Kirk Bateman


2009/8/11 David Knell <dave@3c.co.uk (dave@3c.co.uk)>

[jmesquita at gmail.com]

I am sorry for the ignorance on the matter, but how does google voice does? Do they also have humans?

jmesquita

On Tue, Aug 11, 2009 at 11:17 AM, Kirk Bateman <kirk.bateman@gmail.com (kirk.bateman@gmail.com)> wrote:

[kirk.bateman at gmail.com]

Not sure, but they do certainly have a reasonably large server farm for doing processing

I note that sphinx4 I believe has a java example for doing dictation transcription from an audio file (saw something on a sphinx forum or mailing list while trawling the net).

I'm still investigating modifications to use pocketsphinx.

Regards

Kirk


2009/8/11 João Mesquita <jmesquita@gmail.com (jmesquita@gmail.com)>