I am looking for methods to identify sounds within audio files. I was hoping to find some ready-made user-friendly software to do this but haven't found it yet. I am familiar with Matlab and some C++ and have mech. eng. background so I can do some coding work myself to implement a solution but I'm not a serious programmer.

From my online searches I seem to have found there are 2 potential methods; cross correlation and fingerprinting. It seems to me that the audio fingerprinting method is potentially more accurate than raw cross correlation. At this point this investigation is for my own hobby purposes to be implemented with a robot.

Can anyone with related experience comment which approach may be more successful? It seems there are several audio fingerprinting methods and some open source codes even for matlab so I am intrigued, however, these methods are mostly developed to identify entire songs.

The sounds I want to identify are short (from < 1s to ~5s). I was thinking if the analysis sample rate was increased to fill the short sound file that maybe the method could work with very short files such as these sounds, but I do not know the implications for the rest of the techniques of fingerprinting. Does this seem like a reasonable / possible approach to anyone? If not are there references for other potentially applicable DSP techniques for this I can research?


  • $\begingroup$ Are the sounds to be identified deterministic, or will they vary somewhat each time? $\endgroup$ Commented Mar 21, 2016 at 16:37

1 Answer 1


Audio Fingerprinting can be used to match near identical recordings of sounds with a bit of additive noise. It will not work for similar sounds, cover versions etc.

Without knowing your application, I'd recommend giving Dan Ellis' audio fingerprinting code a try. You'll find the MATLAB and Python versions here:



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.