Voice comparison

[ConsultingJoe]

I was looking to see if I could get a vb script or C++ script that would help me with is.

Prerecord words or phrases then compare back to them

The thing is I don't know VB or C++

I heard theres a way to do it with COM

Please help

[Gene]

I was looking to see if I could get a vb script or C++ script that would help me with is.

Prerecord words or phrases then compare back to them

The thing is I don't know VB or C++

I heard theres a way to do it with COM

Please help

This is a large task, voices are somewhat like finger prints in terms of being relativly unique,there are many variations on a theme. You would have to analyze not the sound files per se, but the wave patterns in the sound streams from both sources separately. Next you would need to compare them as if you had plotted both on pieces of graph paper, put one over the other, then hold them up to a light and slide them back and forth to see if they match. That would be the physicality of it. In the software you would need to record the wave patterns at say every 1/60th of a second for the duration of the sound of interest.

1-- Once you have done that for each sound source, you would average the first 5 points in the second source and compare that average with that of the first 5 points of the original source.

If it matches...

2-- Compare the averages of the next 5 points from each source, and so on, if you match for a second or more that definately be an area of interest.

If it doesn't match...

3-- Then compare the average of the first 5 points of the second source with the 2nd - 6th points of the original source. Then compare the 3rd - 8th points' average from the second source with the 5 point average of the original source stepping through 1 point at a time.

4-- Repeat the steps above all the way through the original source. Whether or not you find matches switch the sources and repeat the steps above. then you will have compared each source with the other.

While doing the comparisons I wouldn't look for exact matches, instead I would start matching if they were within 20% of each other, if that shows that everything matches anything, I would reduce the percentage to 19 and run the test again. Once you get the software to recognize the same word(s) in different recordings of the same person, you can begin to work on recognizing the same word from different people, then you can work on regional accents X as a second language and other issues.

Of course I have assumed in the foregoing that both recordings are from the same person and that no attempt has been made to disguise one of the sources. I have necessarily symplified this in order to finish the response today.

Gene

Edit: Of, course if you or some one else knows of a specific com source for functions to perform the tasks above it would be easier.

Edited February 8, 2006 by Gene