Hi There!

I'm trying to use the fiddle~ object to take a sample and resynthesize it with the partials. Each partial ends up fluttering around and sounding too noisy to use. Is there a way to stabilize each partial and clean up the sound of the final product?

I've attached a link to a picture of the patch with the first 3 partials. Any help is appreciated!

LINK 1: https://drive.google.com/file/d/1rxSEIuOhVbYy_mPAhx6O9OkZfpWBkQKB/view?usp=sharing

I had also posted on reddit and somebody had suggested passing the partials through a rotating buffer and taking the average. It sounds like a good idea, but I'm so new to Pure Data that I'm not sure how to implement it. I found an example using the delwrite~ object to get a delay, but I'm not sure how to get the average over a range of time. I've attached a link to a picture of how the patch looks with the parts I know to add in. Any help is appreciated!

LINK 2: https://drive.google.com/file/d/1FNZLsF7B2s16YUtSHqO9Jlktk7Mv2Bmz/view?usp=sharing

I'm open to any other suggestions as well! I'd love to get this worked out!

I have a moving average abstraction here: https://github.com/jamshark70/hjh-abs/blob/master/moving-avg.pd

Also moving median (requires cyclone as Pd vanilla doesn't have a median function that I could find): https://github.com/jamshark70/hjh-abs/blob/master/moving-median.pd

The moving average is probably faster.

hjh

[fiddle~] provides a "cooked pitch" output with no fluctuations for the (perceived) fundamental.

The 3 partials you are taking from the last outlet are sorted by order of magnitude - not by pitch or partial order- theoretically (one could fluctuate from 1rst to 5th partial in an instant depending on the sound you feed the fiddle.) They update with each dsp block. That is to say, they don't fluctuate at signal rate. A signal rate [~] delay for "rotating buffer" in this case wont work - it will add [dsp blocksize - 1] values of the same for each dsp block and you would need a very long chain of [delread~] to get the values you need to make an average on N pitch values.

SO - If you want to do averaging on those pitch values, you need to set up a message rate (read DSP block rate) rotary buffer. [trigger] and [float] are your friends in this case.

HOWEVER

You are trying to resynthesize a guitar sound. What you need is knowledge of what partials have what amplitude at a given time. Supposing the 3 partials you are getting are #1 #2 and #3 (no guarantee) all you need is to multiply the cooked pitch by the # and gain it with the amplitude outlet

You might want to do some averaging on the amplitude though

Also: You could just slap [line~] on each pitch and amp output for some basic averaging

@bocanegra

Thank you for your response!

The goal with the patch is to smoothly interpolate between the spectra of any two random inputs in real time. I'm very new to Pure Data, so any help or clear walkthroughs are very appreciated!

With the frequencies from the fiddle~ object, there are very obvious jumps between different harmonics, which are pretty brief, but drastically different in pitch. This is exactly what you were describing. I thought averaging the frequency information from a short buffer, then applying a line~ object could help smooth things out.

Before trying fiddle~, my first thought was to slide the amplitudes of each bin of an FFT from input 1 to the amplitude of each bin of input 2, but it still sounds more like a crossfade than a morph.

The fiddle data seems like it could work perfectly for this other than the gurgling pitches of the individual oscillators. If I could somehow hold the frequencies solid with a gate and some kind of attack/transient activated trigger, I think I could get this figured out. Or at least a solid version 1 without sliding or bending pitches.

Again, I'd really appreciate any help, as I don't really know what I'm doing in Pure Data. Thanks again!

I've attached a link to a picture of the current noisy patch: https://drive.google.com/file/d/1ybRgmVSNC30cf5Yb_CpbwylvDD_GELPr/view?usp=sharing

@MCS740 First of all. 0.5 ms wont exactly smooth your output. If you are running at 48khz with a block size of 64, the update rate of your pitch/amp outputs will happen every 1.33333 ms. 1000*blocksize/samplerate is the formula. If you want things to change in a more smooth fashion you need to tell your [line~] object to work slower than that. Also consider that any fluctuation faster than 50ms is perceived by the ear in the tonal range, hence it will be noisy / clicky. Also try @ddw_music 's averaging filter before you add [line~] to the mix

Now about the peaks changing place by the order of amplitude causing pitch to glitch, you can perhaps ameliorate this by running your sample through a lowpass filter, imposing a 3db/octave roll off - and make up for it with a high shelf filter on your final output.

BUT what I was really on about was you don't need to do all that. In order to synthesize your guitar sample all you need is the amplitudes of each partial / envelope of the harmonic spectrum. Fiddle is not ideal for this purpose. Sigmund might be better? FFT is better suited, but will take a completely different (and hairy) approach.

Guru meditation: http://www.pd-tutorial.com/english/ch03s08.html + http://msp.ucsd.edu/techniques/v0.11/book-html/node179.html