Members Hard Truth Posted December 5, 2012 Members Share Posted December 5, 2012 Title: Techniques for Machine Understanding of Live Drum Performances Speaker: Eric Battenberg (http://www.eecs.berkeley.edu/~ericb/) Advisors: David Wessel and Nelson Morgan Date: Monday, December 10, 2012 Time: 11am - 12pm Room: 373 Soda Hallhttp://www.eecs.berkeley.edu/Directions/#soda Abstract: This talk will cover machine listening techniques for the automated real-time analysis of live drum performances. Onset detection, drum detection, beat tracking, and drum pattern analysis are combined into a system that provides rhythmic information useful in performance analysis, synchronization, and retrieval. The talk will focus on the drum detection and pattern analysis components of the system. For drum detection, a gamma mixture model is used to compute multiple spectral templates per drum onto which onset events can be decomposed using a technique based on non-negative matrix factorization. Unlike classification-based approaches to drum detection, this approach provides amplitude information which is invaluable in the analysis of rhythm. The drum pattern analysis component uses a generatively pre-trained deep neural network in order to estimate high-level rhythmic information. The network is tested with beat alignment tasks, including downbeat detection, and significantly reduces alignment errors compared with a commonly used pattern correlation method. Link to comment Share on other sites More sharing options...
Members UstadKhanAli Posted December 5, 2012 Members Share Posted December 5, 2012 We could use a a gamma mixture model to compute multiple spectral templates per drum onto which onset events can be decomposed, using a technique based on non-negative matrix factorization. This would enable us to utilize onset detection, drum detection, beat tracking, and drum pattern analysis, which could then be combined into a system that provides rhythmic information useful in performance analysis, synchronization, and retrieval. We could then employ a generatively pre-trained deep neural network in order to estimate high-level rhythmic information. Or we could say, "Hey, let's get a drummer and and some tube mics and go for it!" Link to comment Share on other sites More sharing options...
Members UstadKhanAli Posted December 5, 2012 Members Share Posted December 5, 2012 I realize that more people are gonna wanna party with someone utilizing a gamma mixture model to compute multiple spectral templates using techniques based on non-negative matrix factorization. We're all wild that way. But once in a while, consider a meat puppet swingin' sticks, just for {censored}s and giggles. Link to comment Share on other sites More sharing options...
Members blue2blue Posted December 5, 2012 Members Share Posted December 5, 2012 Originally Posted by Hard Truth Title: Techniques for Machine Understanding of Live Drum PerformancesSpeaker: Eric Battenberg (http://www.eecs.berkeley.edu/~ericb/)Advisors: David Wessel and Nelson MorganDate: Monday, December 10, 2012Time: 11am - 12pmRoom: 373 Soda Hallhttp://www.eecs.berkeley.edu/Directions/#sodaAbstract:This talk will cover machine listening techniques for the automatedreal-time analysis of live drum performances. Onset detection, drumdetection, beat tracking, and drum pattern analysis are combined into asystem that provides rhythmic information useful in performance analysis,synchronization, and retrieval. The talk will focus on the drum detectionand pattern analysis components of the system.For drum detection, a gamma mixture model is used to compute multiplespectral templates per drum onto which onset events can be decomposedusing a technique based on non-negative matrix factorization. Unlikeclassification-based approaches to drum detection, this approach providesamplitude information which is invaluable in the analysis of rhythm.The drum pattern analysis component uses a generatively pre-trained deepneural network in order to estimate high-level rhythmic information. Thenetwork is tested with beat alignment tasks, including downbeat detection,and significantly reduces alignment errors compared with a commonly usedpattern correlation method. That sounds really cool.No doubt, much of the talk would be over my head, but I love fuzzy logic implementation theory, screw all that binary state stuff. Link to comment Share on other sites More sharing options...
Members veracohr Posted December 5, 2012 Members Share Posted December 5, 2012 How's your gamma mixture and non-negative matrix factorization? It's a solid 43. Link to comment Share on other sites More sharing options...
Members Anderton Posted December 5, 2012 Members Share Posted December 5, 2012 I hit things with sticks. Link to comment Share on other sites More sharing options...
Members UstadKhanAli Posted December 6, 2012 Members Share Posted December 6, 2012 Originally Posted by veracohr It's a solid 43. We always felt your gamma mixture and non-negative matrix factorization was over 40. And you deserve it. Link to comment Share on other sites More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.