Detecting start and end of waves for further processing

I've currently got a program that allows users to record a sample of 3 different beatboxed drum hits (kick, snare and hi hat) and then record a proper beatboxing session of ~10 seconds.

The programs need to compare the frequency content of each beatbox hit in the session with the samples to see if they match up. Here is an example of what the long session looks like:

And here is an example of a recorded sample:

I might be wrong, but I figure that the first thing i need to do is be able to detect the start and end of each hit so that I can compare just that waveform against the samples without any white space.

Is there a way to accomplish this in MATLAB from the editor? I want this to be a standalone program so I'm not using the command window.

If this isn't possible to do, could anyone please suggest a different way of accomplishing my goal?

Many thanks!

Answer 1

[Image Analyst]

How about just thresholding

% Find out which elements have some strong (non-noise) signal value.
soundIndexes = signal > 0.03; % or whatever value you want.
% Now extract out only the elements with sound in them.
croppedSignal = signal(soundIndexes);

[Tom Leverett]

Thanks for your response. I'm very new to MATLAB so I'm not sure how to implement that. Would this be done while it's recording or post-recording? And is 'signal' the name of the recording (e.g. mine is called 'myRecording')?

I'm using the audiorecorder object by the way.

Many thanks

[Image Analyst]

I don't know how to do it "live" while it's capturing data.

Yes, "signal" would be what you call "myRecording".

[Tom Leverett]

That's fine I don't need it to be done real time. When it's finished recording, I want to be able to click a button that then detects each sound wave. So if I put your code into the button callback it looks like this:

function buttonProcess_Callback(~,~)
  soundIndexes = myRecording > 0.03;
  croppedSignal = myRecording(soundIndexes);
end

What's the next step from here?

[Image Analyst]

I guess you'd send your cropped sound pulse through the spectrogram() or pwelch() function. Does that help?

[Tom Leverett]

That's definitely a great start, so thank you for that!

Although I imagine when I add in functionality to compare the samples against the session recording, it wouldn't compare each hit individually, but rather the whole session...

What i want is for it to run a comparison of the sample against points a and b in the full session, where a is the start of the first waveform and b is the end, and do this sequentially for each waveform in the session.

Any ideas on how i might be able to detect points a and b etc.?

Sorry to be so demanding of you!

[Image Analyst]

Not really sure what to do as I'm not an audio engineer. I don't know what "compare" means when you say "compare the frequency content of each beatbox hit". The spectrogram(), pwelch(), and fft() will give you the frequency content, but what to do with that to do the "compare", I have no idea. I'm not even sure you do. How would you want to compare them. Of course they'll be different , but how are you going to assess the amount of difference and what are you going to do based on that knowledge?

[Tom Leverett]

My apologies, I haven't explained the full idea of my project. What I'm essentially trying to do is create a program that converts beatboxing into a MIDI track. In order to achieve this, I figured that the best way was to get the users to first record a 'hit' of each drum and after that record a ~10 second long track that will contain a mixture of those hits (kick, snare and hi hat).

With these recorded, the program would compare the frequency content of each waveform in the long track with the sample hits to see if they match up (~80% accuracy would be accepted). The program will then use this data to generate a MIDI track with the right drum types in the right places.

Anyway, as you've said, you don't know how it might work, but hopefully someone else might see this!

[Image Analyst]

So you want to recognize if a sound is in your sound library, and if it is, replace the actual sound with the "perfect", or at least standard , sound from your library?

[Tom Leverett]

Not exactly. I want to recognise if a sound is similar to another that the user records, and if it is, then I want to get the relevant data to generate a MIDI track where the notes have all the right parameters (Starting point, duration and pitch). To visualise, it's basically using this:

to make one of these:

where all of the notes are in the right place and are the right pitch.