We propose a sound recognizer that uses a reduced featureset to identify musical instruments from single notes in sound recordingsas well as the note that was played. The recognizer learns a set of spectralfeatures from the data using non-negative matrix factorization.The accuracy of the recognizer is very high for both instrument and noteclassi cation: the recognition rate for instruments ranged from 94% to100%, while for note identi cation ranged from 86% to 100%.
|Title of host publication||Proceedings of INForum|
|Publication status||Published - 1 Jan 2011|
|Event||INFORUM - |
Duration: 1 Jan 2011 → …
|Period||1/01/11 → …|