Adaptive onset detection based on instrument recognition

摘要

Onset detection is the foundation and key to high-level audio processing like music retrieval and transcription. Research shows that the detection algorithm is associated with instrument category, and high accuracy can be achieved in instrument recognition studies. Thus the adaptive detection system based on instrument recognition was proposed in this paper. The system uses HMM classifier to identify input audio falling into four categories, adaptively adopts suitable detection algorithm for each type, and output onset times in the end. The experiment results show that onset evaluation values, such as the F-measure value, have been improved in the system.