[60fps] Grzegorz Kwiatek (Greg) - "Misunderstanding" -- Chiptune Visualization / Atari SAP

** PLEASE, WATCH IN HIGH QUALITY FOR PROPER AUDIO! **
So, I decided to take a look at all the separate channels of some chiptune music at once. Thought it might look cool, since those waveforms are usually so pretty.
Here's Greg's "Misunderstanding" once more, but in 60 FPS and with the new TimbreCapture algorithm.
This is my favorite Atari POKEY chiptune, but it has a bug: if it loops, the tempo of the loop is reduced to 1/4, which gives this song a very somber and ominous feel to it.
However, that is not how it was intended to be played, apparently, so true to the name that was nothing but a misunderstanding from my part. I still love that bugged version and to me that'll always be the real version of this song. This video with 7:03 minutes displays the slowdown bug in the second part.
This is a video made with a complete rewrite of my original Python script. I finally got around to implementing my "TimbreCapture" algorithm, which finds the exact waveform shape being repeated for a note (the timbre's minimum waveform) and consistently centers it on screen. This way we can always see the exact shape of the wave being played at any given moment and how it changes.
It's working well so far (misses a few weirder waveforms), but I still need to add a couple more features to it.
Oh, and there's more! I'm also going to be releasing my tool to convert POKEY dumps to MIDI files. These will require a modified version of asapscan from the ASAP library.
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
The Grayscale Project is an awesome Atari 8-bit music collaboration from Poland. Members are:
Grzegorz Kwiatek (Greg), Łukasz Sychowicz (X-Ray) and Bartek Wąsiel (BeWu).
grayscale.scene.pl/en_index.php
You can find their work, along with tons of other great Atari music, on the Atari SAP Music Archive (ASMA) at:
asma.atari.org/

[@hashtagrich]

Still one of my favorite musicians of all! :)

[@SyncdAlien]

Loving it! 😍😍 Subscribing to you now! 😍

[@lisumax111]

new algorithm looks so dynamic

[@1ucasvb]

Yeah, it's working really well on this track. It still fails badly in a few cases. I'm trying it out on some FM sounds from Sega Mega Drive/Genesis games and it's not very good in those cases. My heuristic of using zeros of the waveform is not as robust as I thought, so I'll try matching waveforms based on their signatures of the peaks, valleys and zeros.
I'm hoping to figure it out for completely arbitrary timbres (like, an actual violin or piano recording) so we can all use it on anything we want.

[@lisumax111]

Violin or piano? I believe that chiptune looks so good with such visualization because it is defined and created by simple mathematical formulas (or whatever you say in English), or at least simpler than sound of a violin. Its would look rather chaotic when (or just as a regular visualization) when playing natural sounds.

[@1ucasvb]

The simple thing about these chiptunes is that the timbres don't change in time very much because the harmonics are sharp and the entire spectrum decays at the same time, so the entire shape of the wave just shrinks.
If you play a single note on the piano and hold it, the harmonics and overtones decay at different rates. This means the "local timbre", the little bit of wave that is characteristic of the sound, changes shape until it's almost a pure sine wave.
This change can be quite dramatic for different instruments and I've made some visualizations of this in the past. It works fine, as long as you have a way to "lock in" the timbre shape consistently. This is easy to do manually when you know beforehand what note you are playing and when it started, but I want to figure how to do it in general. This is the key idea I want to address in these visualizations, and it is not really restricted to chiptunes.
As long as you are looking at a single note being played, if you zoom in on the waveform it will always look like a really well defined pattern. Here's an example of a violin (top) and piano (bottom) both playing an A5 (at 880 Hz): i.imgur.com/stMOXkH.png
When you have chords and multiple instruments then the whole thing is an incomprehensible mess. So this is only useful if you can isolate every instrument and every note, which is possible in the case of the chiptunes or if you are creating the music yourself (which is what I'll be doing in the future).

[@Formula_Zero_EX]

Sounds like a jungle level in a shoot-‘em-up.

[@jonathonbreitner1334]

Are you able to run your Lazy Discrete Fourier Transform in this program, It would be so freaking cool to see the frequency domain of each track too !

[@1ucasvb]

No reason to use LDFT here, though. FFTs would do just fine.
And I've looked into it, but these waveforms look really dirty and messy in frequency domain. Pure square waves are nice, but these weird duty cycles really mess up everything.