Isolating specific phonemes

Hello everyone!

I am fairly a beginner in the audio processing field.
I have searched the forums and played a little with some of the plug-ins already developed, but I haven’t found something that could help me.

My problem is that I have thousands of recorded words from witch I have to isolate a specific phoneme : ‘r’.
This is very much time consuming, doing it manually, but my greatest problem is the accuracy of the isolation. Because this is only the starting part of my project. Later I will want to use those isolated parts to identify them in other words. That is why I need the isolated parts to belong only to that specific phoneme.

If anyone could give me some pointers to speed up the process or to make it more accurate, it will be greatly appreciated.

Thank you!

What’s the actual job? Sometimes we can think of solutions by changing the job a little.

You probably didn’t get up this morning and say: “I think I’ll go isolate some phonemes.”

Koz

The job is actually separating the ‘r’ phoneme from those words. I need it in order to build a database that will help me find occurrences of that phoneme in other recordings.

My main problem is the fact that is very difficult to identify the boundaries.

Carnegie Mellon’s Sphynx can do that. It isn’t easy and will require some experimenting. Phoneme recognition is a lot harder than word recognition. See:

https://sourceforge.net/projects/cmusphinx/

And a page from CMU about the subject of phoneme recognition:

http://cmusphinx.sourceforge.net/wiki/phonemerecognition

It’s a free open source program, but I think you’ll have to compile it yourself. libraries for different languages are available.

Thank you cyrano!

I will test them and return with a feedback.

Unfortunately I cannot seem to find my way around Sphynx .

Maybe someone could point me to an audacity feature or patch that could help me with my work?
I’ve already tried PhonemeFinder.ny and Segmentation.ny.

Thank you.