More fun results! I upgraded the snippet matching procedure to use a weighted combination of spectral centroid distance, spectral peaks distance, and tempo distance. The results are becoming much more coherent. I have the beginnings of an emacs-based interactive looper UI as well, so basically this could become an improvisational rhythm composer. Here are the sounds: http://xelf.me/resource-55.mp3
On Sun, Dec 20, 2020 at 6:34 PM David O'Toole <[email protected]> wrote: > Happy Holidays everyone! > > It's been an amazing adventure. I now have a basic corpus-based musical > concatenative synthesis system written in S7 Scheme on top of Snd's audio > tools. It analyzes a bunch of input files, splits them into tiny FLAC > fragments sliced configurably on the beats, and writes the analysis data > for each song (with all the slices) in its own file. Later you can load any > selection of these data files, and run metadata searches to find matching > slices. Only the slices that you actually end up choosing get loaded, so > the searches themselves are very fast. Right now I've been feeding it dance > and hip-hop music and doing matches based on minimizing distance between > the loudest spectral peaks in each chunk. Here are some results: > http://xelf.me/scheme-mosaic-2020-collection-1.mp3 Most of these use a > chopped-and-screwed backing track (also done with Mosaic) and matching > segments of other songs to them. > > The next step is to work more on my interactive looper so that one can > "jam" modifying the loop as you go. > > The homepage is http://xelf.me/scheme-mosaic.html > > After the success of this year's Emacsconf 2020, I've started the EMSIG > (Emacs Multimedia Special Interest Group) covering such tools as Snd, > Ecasound, EMMS, Midi, etc. you would be welcome to visit the irc channel > #emsig on freenode or email me if you are interested. > > Thanks everyone (especially Bill!) for your help in solving little > stumbling blocks on the way. >
_______________________________________________ Cmdist mailing list [email protected] https://cm-mail.stanford.edu/mailman/listinfo/cmdist
