Comparing Audio Files with Python

Posted by Agreeable-Bluebird67@reddit | learnprogramming | View on Reddit | 2 comments

I’ve been using librosa and sound file for some basic metadata retrieval info Python, but would like to expand to automate comparisons between short audio clips. What other libraries or functions inside librosa would be best to analyze material like drum samples? Is there a way to identify the source of the sound (kick, snare, tom) without training my own machine learning algo?

[-]

dmazzoni@reddit

Distinguishing between different major types of drums should be relatively easy using librosa.

librosa.pyin - will give you the fundamental frequency, which should be low for a kick, medium for a tom, and not useful for a snare.

librosa.feature.mfcc - the first few coefficients encode some good general information about the spectrum:

[0] - overall energy / loudness

[1] - overall spectral slope (more high-frequency heavy vs low-heavy)

[2] - spectral curvature (how peaked it is)

I suspect mfcc[1] would be good at detecting a snare drum

Try printing out those four values for a bunch of drum samples and see if you can come up with a way to identify them using those numbers.

Agreeable-Bluebird67@reddit (OP)

Wow this is incredibly helpful. Thank you very much for the detailed response