Finding the characteristics of signal

Hello everyone, i am a final year student at my local university and i would like to ask something that is related to my final year project. Right now, i have a sample of an audio file that contain three types of drum instruments which is snare, kick pedal and cymbal. From the attached file, you can see that there a different between the wave so here are the sequence of the wave, the first 15 wave are the snare, the second 16 wave are pedal kick, the third 13 wave are snare and the last 15 wave are cymbal. So, from these audio sample, how can I differentiate the type of the drum instruments ? is there any characteristics of the drum instrument types that make it difference from the other eg: frequency or wave length?

Answer 1

[Star Strider]

I would use the spectrogram (link) function to see what the differences are, then go from there. There are relatively sophisticated methods to separate the different instruments, the independent component analysis for example using the File Exchange contribution PCA and ICA Package (link), and others.

[faiz hazizi]

okay sir, thank you so much. For the spectrogram function, is it need for me to convert my drum audio sample into the spectrogram ? and there are several example of the spectrogram function on the link that you provide. Which example should i follow ? because i did not familiar with the spectrogram. Your help is much appreciated.

[Star Strider]

My pleasure.

I would begin by using the the default parameters, and letting the function create the plot:

spectrogram(YourSignal)

Then change the input arguments (and request the outputs), depending on what you want the function to do, and how you want to analyse your signal. You will have to experiment.

[faiz hazizi]

here are the spectrogram of the whole drum instruments. i don't really understand what is the spectrogram is sir. sorry sir, but from the spectrogram how am i gonna see the difference between those instruments ? eg: from the spectrogram how can i see the difference of snare, kick and cymbal ? and one more thing sir, what do you mean by "Then change the input arguments (and request the outputs), depending on what you want the function to do" ?

[Star Strider]

I would decrease the overlap, use a shorter window, and a larger value for ‘nfft’ (consider at least 2^10, since larger values increase the frequency resolution of the Fourier transform) to begin with.

‘what do you mean by "Then change the input arguments (and request the outputs), depending on what you want the function to do" ?’

I meant exactly that. The spectrogram function has a number of parameters (arguments) you can change so the function will do what you want. I suggested three already.

Please spend some time reading the documentation for the function to understand what the various arguments do. You have to experiment with them to get the result you want. Remember, you want to identify the instruments by differences in their spectra, and this requires increased frequency resolution and time resolution.