similarity and segmenta on for electronic dance music
play

Similarity and Segmenta.on for Electronic Dance Music Bruno - PowerPoint PPT Presentation

Universiteit van Amsterdam 17 January 2013 Similarity and Segmenta.on for Electronic Dance Music Bruno Rocha Aline Honingh Niels Bogaards Victor Bergen Henegouwen Goal Similar


  1. Universiteit ¡van ¡Amsterdam 17 ¡January ¡2013 Similarity ¡and ¡Segmenta.on ¡for ¡Electronic ¡ Dance ¡Music Bruno ¡Rocha Aline ¡Honingh Niels ¡Bogaards Victor ¡Bergen ¡Henegouwen

  2. Goal Similar tempo Similar timbre Electronic Mobile Dance Music App (EDM) Similar rhythm ...

  3. Methodology Seg B2 Music Timbre Rhythm Features Features Electronic Dance Music Distance User Collection Seg A1 Timbre Rhythm Song A Song B Features Features Seg A1 Seg A2 Seg A3 Seg B1 Seg B2 Distance Song N Seg A3 Timbre Rhythm Features Features

  4. SimMixer iPad test app for timbre similarity • database contains 1295 EDM songs • segmented into 22745 segments • each segment is represented by 31 timbre descriptors • SimMixer is a tool to explore and validate the timbre similarity algorithms

  5. Research Issues Segmentation • Detection of first downbeat • typical EDM structure starts with bass drum • critical step for the detection of the structure Figure 1: Detection of first downbeat

  6. Research Issues Segmentation • Segment boundaries fine-tuning • segmentation algorithm detects timbre changes • segment boundaries in EDM are known to usually coincide with beats Figure 2: Segment boundaries fine-tuning B. Rocha, N. Bogaards, A. Honingh, “Unsupervised detection of structural changes in electronic dance music”, DMRN, UK, 2012

  7. Development • Work method • develop a working algorithm • test feasibility • test usability • develop product(s) Matlab C++ iPad optimized Figure 3: Tool Chain

  8. Lessons Learned • optimized implementation can only start after the algorithm is done (lost quite some time implementing sub-algorithms that in the end are not used) • important to establish ground truth / reference early on, but a problem is that it may not be clear what exactly needs to be researched / developed (for instance: problem of segmentation) • plan research sub-projects carefully, to allow implementation to start asap and increase involvement by industrial partner

  9. Conclusion • University vs (small) Company • Science vs Commerce • Science - Company - Audience

  10. Horizon • Intelligent mobile music tools

Recommend


More recommend