I'm trying to learn more about voice recognition, and I was looking this library here and noticed there were two implementations of MFCC (one called mfcc and another called mfcc2). I ran both of these on the same data and got different results, and was wondering why there are two different version of MFCC that give two different results. Thanks.
At a glance - the number of filters is different, the first algorithm has it equal to 48, while the second one - to 32. But even these small things aside, MFCC computation is basically a pipeline of algorithms with 4 or 5 stages. And each stage has some input parameters varying which you can get slightly different output. Given that each stage is implemented properly these differences have negligible effect on the recognition result.