Hi everyone,

I'm faced with a challenging problem involving rhythm detection for brass and wind instruments.

If a saxophone (or similar instrument) plays four quarter notes continuously of the same pitch (let's say C), what do you think would be the best way to detect the start of each note— specifically if these notes are played smoothly with no clear separation between them?

When the notes are clearly articulated (with a silence between them) I have been using sigmund~'s notes parameter to detect the attacks, but when the note is a continuous stream with a very subtle attack this does not work so well.

I have also experimented with bonk~ but I can't seem to have any luck setting the parameters to ensure that it doesn't give "too much" output and detect attacks that aren't there or are coming from different sound sources.

Any tips would be greatly appreciated, thanks!

@chmod Hi, have you tried using sigmund's pitch parameter instead of notes? Even if the consecutive notes are the same pitch, there should be a slight pause in the audio between notes, which the pitch parameter seems to spot.

If the reported pitch is not accurate enough, you could use both pitch and notes arguments, and then use the -1500 output from "pitch" to separate the output of "notes".

@weightless that's a great idea, I hadn't thought of that, the fact that sigmund outputs -1500 during silences in pitch mode.

I knew I had to use a combination of object in some way, thanks a lot!

@weightless

I tried using that method today and found that if the notes are "tongued" and there is still a continuous pitch between every articulation, the -1500 does not appear as needed and the individual attacks are not picked up. This was especially apparent when I tested using my voice (saying "laaalaaalaaalaaa" for example)— there is a continuous pitch being picked up by sigmund~ but I am articulating the note four times.

I've been trying to look at sudden jumps in the envelope instead but I need to find something that works "one size fits all" for different kinds of instruments and different volume levels as well, so it's been pretty tricky.

@chmod Yes, that makes sense. The problem is that if there is no silence to speak of, and the pitch stays the same, how can we know that there is a new note being played?
I would think that, even if the consecutive notes are tongued, [sigmund~ pitch] would detect some crazy pitch for a few milliseconds during the transition, however smooth that might be (which would be the Ls in your "lalalaaaa" example). So perhaps taking an average of the last n pitches from [sigmund~ pitch], and try to spot the transitions whenever there is a sudden jump in the average? Or something like that.

That being said, I don't think it's going to be so easy to get a universal method that works all the time, but sigmund is the best we've got in Pd for this kind of stuff.

If you don't mind sharing a couple of samples of some particularly critical sources I can do some experiments as well.

@weightless Hi again —

I was trying to look at the envelope of the signal using env~ and subtracting this output with a delayed version of itself to find out where the envelope rises and falls. This seems to work pretty well for most of the instrument samples I was working with but I ran into a problem with a flute recording with a pretty heavy vibrato. Each "vibration" of the vibrato was detected as its own attack since thats where it would peak in the waveform as well.

Your suggestion seems like it could work — I have all of my audio files in an Ableton project right now, is it alright if I send it to you that way (with all the samples in the project folder that is)?

@chmod You should also check out Helmholtz...... http://katjaas.nl/helmholtz/helmholtz.html
It uses one of the most (currently available) advanced techniques. Maybe [Sigmund~] uses similar methods though?
Anyway, the explanation is a very good bed-time read..!
@katjav 's site is great for everything FFT and Pd.
David.