Open tempo, key, and waveform analysis for your library
MCLD 2012 | Sun 09 Aug 2 p.m.–2:45 p.m.
Presented by
-
Pardis Noorzad
@djpardis@mstdn.business
https://djpardis.com
Pardis Noorzad is an engineer and researcher. Pardis created Cueport, a local-first Mac and iPhone music player for DJs, and Trail Mix, its audio-analysis toolkit. Pardis built B2B data infrastructure at General Folders and led data teams at Carbon Health and Twitter. Pardis holds master's degrees in artificial intelligence and applied mathematics.
Pardis Noorzad
@djpardis@mstdn.business
https://djpardis.com
Pardis Noorzad is an engineer and researcher. Pardis created Cueport, a local-first Mac and iPhone music player for DJs, and Trail Mix, its audio-analysis toolkit. Pardis built B2B data infrastructure at General Folders and led data teams at Carbon Health and Twitter. Pardis holds master's degrees in artificial intelligence and applied mathematics.
Abstract
Cueport is a local-first music player music you own. For DJs who use Serato, it imports the tempo, key, cues, and beat grids Serato has already calculated. Listeners who are not DJs, and don't use platforms like Serato, have no analysis to import, and serving them meant building an analyzer. trail mix is that analyzer, an open source Rust toolkit that estimates tempo and musical key and generates compact waveform data, all on the local machine. This talk shows how the analysis works, and explains how single-tempo accuracy is evaluated separately from multi-section tracks using representative test music, reproducible benchmarks, and heuristic confidence scores.
Cueport is a local-first music player music you own. For DJs who use Serato, it imports the tempo, key, cues, and beat grids Serato has already calculated. Listeners who are not DJs, and don't use platforms like Serato, have no analysis to import, and serving them meant building an analyzer. trail mix is that analyzer, an open source Rust toolkit that estimates tempo and musical key and generates compact waveform data, all on the local machine. This talk shows how the analysis works, and explains how single-tempo accuracy is evaluated separately from multi-section tracks using representative test music, reproducible benchmarks, and heuristic confidence scores.