Pitch to midi in SM?

nix · 7 Mar 2008

Hey again, To start with, I was wondering if it's possible to convert a float value to an integer value? Help would be appreciated muchly. Cheers, nix

Acrobat · 7 Mar 2008

nix wrote: Hey again, To start with, I was wondering if it's possible to convert a float value to an integer value? Help would be appreciated muchly. Cheers, nix

..in green data, just connect an F box to an I one, in poly audio, you need to round integer in code...

rl · 7 Mar 2008

> trying for pitch to midi in SM...

I'm not sure what you need. The pitch output of the Midi2Poly component carries the MIDI note number.

oddson · 7 Mar 2008

I think the idea is to create a midi stream from audio pitch data.

nix · 7 Mar 2008

Thanks very much Acrobat, That really helps. The problem I am having now is that I'm triggering multiple note on and offs, rather than just 1 when I introduce the audio signal. I'll try again now. That's what I'm trying to do oddson. If it could be accomplished you could even use your voice as a midi instrument. Wishing you satisfying creations, nix

oddson · 7 Mar 2008

Re: Pitch to MIDI In theory... you could filter the input to a passband that contains the fundamentals you want to capture. Then periodically 'sample' small bursts of audio and run them through the FFT component... use the max value in the array to determine the dominant frequency and calculate a pitch-bend or new note based on that...

Whether to use pitch-bend or new note would be based on the output of an envelope follower that's tracking the dynamics of the input - as would note-on and off events and their velocity.

Hurdles: The current FFT component may not have the resolution required. The frequency of 'sampling' and processing the results might be way too slow to be usable. The signal would have to be pretty clean - and intrinsically monophonic (in the sense of 'one note at a time').

As I understand it these schemes - when done by professional DSP programers - have always been a bit hit-and-miss. Even trying to do a pulse follower in SynthMaker I had pretty unsatisfactory results. I'm guessing this is an idea whose time has not-yet come.

rl · 7 Mar 2008

you could also try the poor man's approach to pitch estimation. High-pass filter the signal and count the transitions through zero. This will of course introduce some latency e.g. with a base frequency of 100 Hz the signal has a base period of 10 ms

oddson · 7 Mar 2008

That's likely to be a more productive approach given the limitations of the program... I might just try it... thanks Rudolph!

angstrom · 7 Mar 2008

I have an MS20 which uses this type of pitch tracking, although it's not very accurate it has a certain sound which is quite fun. 'Wailing' or 'electronic yodeling' might describe it !

I would go for a lowpass rather than a hipass, you want to try and keep the fundamental and discard everything else.

oddson · 8 Mar 2008

I'm sure he meant low-pass that's how I read it, I didn't even notice is said 'high'. It should be a bit of a brick wall too. I think the real problem is the possibility of multiple crossings where the fundamental approaches zero. So a steep rolloff would be essential.

Then the trick would still be to get the dynamics tracking working well enough to be passible. You'd have to hold off trigging the event until the new pitch had been detected. And you'd still have to sense pitch-bend vs. new note.

nix · 8 Mar 2008

Hey again, Thanks very much for your help. Both of these methods seem beyond me, but I will go over what's been said again. I am still interested to try. To take another approach, is there any way to control an oscillator without using midi, just audio? I know of FrettedSynth Audio's SE guitar controlled synths, and would hope to achieve something similar. If you would make the zero crossings method, oddson, I would be curious to try it. Wishing you satisfying music, nix

oddson · 8 Mar 2008

I did a little investigating and found the zero crossing method is not very reliable even in theory. Unless you strongly filter your signal to within a single octave band you can easily get multiple crossings from a signal with even moderate amounts of low-order harmonic content.

I was thinking if you summed the (absolute) values of each sample since the last crossing you could have some kind of threshold check that tries to determine if it is a localized crossing or the result of the fundamental oscillation. But it's just a thought so far and it would have to be calibrated with changes in the overall volume of the signal...

Anyway... I'll give the basic version a try and see where that takes me.

oddson · 8 Mar 2008

streamin x; streamout f; float x1,zx,i,i1,w,w1; // x1 is x @ t-1 // zx a mask -> upwards zero crossing is true // i is counter of cycles since last zx // i1 is the increment i is based on // w is the current wavelength calculation // w1 is a lowpassed filtered version of same

zx = (x1 < 0) & (x >= 0); // true when upward zero crossing reached

i = i1 - (i1 & zx);

w= (zx & 1 ==0 & w) + (zx & i1);

hop(16){ w1 = 0.25 * w + 0.75 * w1; f = 1/w1; }

//lagged/incremented values i1 = i + 1; x1 = x;

This code will extract the normalized frequency of a signal based on upwards zero crossings.

I've made no investigation on using different hop values for the wavelength averaging (using a low-pass filter on the calculated integer values) or on the responsiveness to the output frequency calculation nor have I given much thought at all to the averaging process or the lowpass coeff. It's just a rough sketch for now.

I should have been easy but I'm still not very good at masks. It takes me ten times as long to figure out how to do something with them as it would with normal if-then logic. 😕

oddson · 8 Mar 2008

The results are as expected... simple waveforms read OK but others less so. 🙁