I checked it more thoroughly and it came out that exactly 50% of sounds are extracted.
If you ordered the sounds 1, 2, 3, ..., n the program would produce only every second sound in the following order:
1: sound no n-1
2: sound no n-3
3: sound no n-5
....
n/2: sound no 1
This may not always be exactly the case as I tested it with the file that had an odd numer of sound notes (39). If the number is even it may be slightly different. But you get the general idea.
I also downloaded the source code and spotted some
occurences there. I am not a programming expert however so maybe it has absolutely nothing to do with it. I' m thinking of experimenting with the code myself a bit.