HAMR@ISMIR 2014 Proceedings

====== Differences ====== This shows you the differences between two versions of the page.

--- spectral_clustering_for_pattern_discovery [2014/10/26 04:50]
urinieto created
+++ spectral_clustering_for_pattern_discovery [2014/10/26 05:19] (current)
urinieto
@@ Line 1: / Line 1: @@
-====== Spectral Clustering for Pattern Discovery ======
+====== Spectral Clustering for Pattern Discovery (The McFee Hack, Part 2) ======
 | Authors | Oriol Nieto, Brian McFee |
 | Affiliation | New York University |
 | Code | https://github.com/urinieto/laplacian_segmentation |
+===== What =====
 We modified the [[http://www.terasoft.com.tw/conf/ismir2014/proceedings/T073_319_Paper.pdf|Spectral Clustering]] algorithm such that it identifies musical patterns (motives, sections, segments) of a piece.
+===== How =====
+It works like this:
+  * Obtain segments from a small scale layer (e.g. m > 8) using spectral clustering.
+  * Filter segments by keeping the repeated ones only (by using the label information).
+  * Repeat for N layers
+  * Aggregate patterns, because overlap is allowed in pattern discovery.
+Spectral Clustering on Come Together by The Beatles for m layers:
+{{ :screen_shot_2014-10-26_at_4.50.14_pm.png?700 |}}
+===== Results =====