PDA

View Full Version : Sound file to speech transcript library?



nickname_changed
12-16-2004, 10:55 AM
I've not given much thought to this before really, but has anyone come across a library or program that can take a sound file (say an MP3) and extract all the speech (a sort of transcription). Whether it is grammatically correct or anything like that I don't care, I just want it to list all of the words used so I can search through them later.

The reason I'm asking is when I go to uni I would like to record all of the lectures etc and store them in a database, and later in the year to be able to search through all the recordings for information on a particular subject.

Any information/resources/links would be great! If I can't find a library (C++, COM, .NET, any) I'll attempt to write something myself.

Thanks, and if I can throw something together I'll definately give it to anyone thats interested.

Sang-drax
12-16-2004, 11:18 AM
I doubt it.
Some programs offer speech recogition, but they cost money and they need to be trained a lot before they become any good.

doubleanti
12-16-2004, 12:26 PM
>I'll attempt to write something myself.

DIY! =)

It's really fun stuff to try to get into, which is why I'm doing it for research this year. You get to learn your DFT's in and out (which you need anyway if you're going into EE) and it's really neat. But it requires linguistics! Read read read... I'm trying to do it right now, just gotta get past all the mathy DFT stuff first, almost at a nice, evenly distributed (in frequency) spectrograph.

Pick up any fundamentals of speech processing book to learn about formants, and formant movement and pick up an intro phonology book to learn about the IPA. It's so fun!

Anyone doing anything remotely interdisciplinary excites me, especially speech processing!

Perspective
12-16-2004, 12:50 PM
go to http://sourceforge.net/search/ , type in "Speech recognition" and check out the source project of your choice.

anonytmouse
12-16-2004, 02:47 PM
Check out HP Speechbot (http://speechbot.research.compaq.com/) for what can be done in this field. Also try a search on speech recognition library (http://www.google.com/search?q=speech+recognition+library+open-source).

Lurker
12-16-2004, 04:25 PM
Do you want the program to recognize only english sounds, or many different sounds? There are quite a few sounds out there, so you may want to try only english first.

Doubleanti: I tought myself how to pronounce ejectives today :) . I'm happy happy.... Love that s'!

doubleanti
12-16-2004, 05:50 PM
Haha, good job~! ::pats you on head::

I assume you mean non-English ejectives haha, else I'd say of course you knew how to... =)

Go for a Cambodian injective now, enjoy! haha...

nickname_changed
12-17-2004, 09:54 AM
Thanks for all the links guys, I found a few librarys that may do what I want, but they all seem to be in Java :P

Lurker: yep, just English sounds and words. I don't even need it to recognise punctuation, if it just outputs something like:


hi john how are you i am well thanks peter today we will be talking
to you about the wonderful work being conducted by MIT into
the benefits of crop rotation for coconut farmers


Just so that I have some list of keywords to seach on. Then I can list the closer looking matches.

Will let you know how it goes, thanks for the links!

anonytmouse
12-17-2004, 11:09 AM
Sphinx (http://cmusphinx.sourceforge.net/html/cmusphinx.php) 2 & 3 are coded in C.