home studios equipment staff/friends booking/rates for sale forum contact

PRF Monthly Tribute Series 40: ABBA

Moderators: kerble, Electrical-Staff

Re: PRF Monthly Tribute Series 40: ABBA

Postby b1l-paraquat on Wed Nov 01, 2017 11:37 pm

hench wrote: i think i'm hearing the chord changes for "dancing queen" -- did the program take pitch into consideration, or is this just musician brain hearing patterns after being told that the timing will work? anybody who writes a dang program for this thing gets a vote, that's for dang sure!


Yeah, it followed the chord changes pretty well, I thought. It does take the fundamental frequency into account for every slice, matching by pitch and octave. Then it tries to pick a best fit from all available matches, the best fit having the smallest sum of differences between all measured frequencies.
It’s pretty fun, but needs more work— at the moment it’s a challenge to create things that don’t sound totally garbled; I remodeled Satie’s Gymnopedies 1 using the same Sabotage data set and while the result sounded nothing like the original, having something to model after gave structure to what would otherwise be random sounding snatches of drum, guitar and voice.
l_o_c wrote:I couldn't find any quotes on Sun Ra's political affiliation, beyond the need for there to be a Black House, and that the White House should be sent to the moon. I'm unaware if these are current Republican issues.
User avatar
b1l-paraquat
viscount
viscount
 
Posts: 34
Joined: Tue Apr 04, 2017 12:05 pm
Location: Effingham, IL

Re: PRF Monthly Tribute Series 40: ABBA

Postby b1l-paraquat on Wed Nov 01, 2017 11:46 pm

A word about “mr tyzik - arrival”, Brian:
I enjoy the aesthetic here a lot—It reminds me a little of Scorn, and by that I mean the *good* Scorn, when it still sounded dangerous and still warm from the Napalm Death oven. I’m a fan
l_o_c wrote:I couldn't find any quotes on Sun Ra's political affiliation, beyond the need for there to be a Black House, and that the White House should be sent to the moon. I'm unaware if these are current Republican issues.
User avatar
b1l-paraquat
viscount
viscount
 
Posts: 34
Joined: Tue Apr 04, 2017 12:05 pm
Location: Effingham, IL

Re: PRF Monthly Tribute Series 40: ABBA

Postby the finger genius on Thu Nov 02, 2017 4:47 am

paraquat, that's awesome.
jimmy two hands wrote:Then we all log off internet and eat nachos. All is well.

Eighteen Hundred and Froze to Death
User avatar
the finger genius
took picture of naked guy
took picture of naked guy
 
Posts: 1187
Joined: Sat May 31, 2008 10:04 pm
Location: NJ

Re: PRF Monthly Tribute Series 40: ABBA

Postby Frank Tuesday on Thu Nov 02, 2017 7:53 am

b1l-paraquat wrote:
Yeah, it followed the chord changes pretty well, I thought. It does take the fundamental frequency into account for every slice, matching by pitch and octave. Then it tries to pick a best fit from all available matches, the best fit having the smallest sum of differences between all measured frequencies.
It’s pretty fun, but needs more work— at the moment it’s a challenge to create things that don’t sound totally garbled; I remodeled Satie’s Gymnopedies 1 using the same Sabotage data set and while the result sounded nothing like the original, having something to model after gave structure to what would otherwise be random sounding snatches of drum, guitar and voice.


This is fascinating. Apologies if you've already answered these: Does it read from a midi file, then try to find clips that match that match the midi note? Or is it reading the song from an audio file? On your Sabbath data set, did you have to manually split the clips, or does the program do that?
Brian (aka Frank)
Math Patrol
User avatar
Frank Tuesday
sister
sister
 
Posts: 52
Joined: Sun Nov 29, 2015 8:14 pm
Location: Austin

Re: PRF Monthly Tribute Series 40: ABBA

Postby b1l-paraquat on Thu Nov 02, 2017 9:57 am

Frank Tuesday wrote:This is fascinating. Apologies if you've already answered these: Does it read from a midi file, then try to find clips that match that match the midi note? Or is it reading the song from an audio file? On your Sabbath data set, did you have to manually split the clips, or does the program do that?


The program works straight from wav files. You pick a wav file for the program to model (or imitate), then pick your wav file "sources" to imitate with. For every wav file, it chops the sound data into pieces of a sample size that you specify, which is a power of 2. The pitch analysis is all done in-memory, prior to processing, so you could probably build and output raw midi information from a wav file-- but to answer your question, no you don't need to feed anything midi into it for it to work.

The frequency and pitch analysis is fast, what takes time is comparing the spectra of every pitch-matched section of sound in the sources to every section of sound in the model. Fortunately, you only have to do this once--for every model-to-source pairing, it saves the result to use later when you're ready to "remodel" and export to wav.

I got the basic idea a couple of years back from a dissertation written by Tristan Jehan. He called the process "Musical Cross-Synthesis". http://web.media.mit.edu/~tristan/phd/dissertation/chapter6.html#x1-950006.5.
If you have an hour, Brian, his dissertation is a great read/skim-- a lot of it documents the different capabilities of a program that he wrote all through school called "Skeleton". Unfortunately, he's not releasing any of his code; I finally just had to sit down and try and replicate it to see what it would sound like.
l_o_c wrote:I couldn't find any quotes on Sun Ra's political affiliation, beyond the need for there to be a Black House, and that the White House should be sent to the moon. I'm unaware if these are current Republican issues.
User avatar
b1l-paraquat
viscount
viscount
 
Posts: 34
Joined: Tue Apr 04, 2017 12:05 pm
Location: Effingham, IL

Previous

Return to General Discussion

Who is online

Users browsing this forum: Google [Bot] and 10 guests