Title :
Real-time music synthesis by patch-based method
Author :
Zhu Bin ; Wang Xin
Author_Institution :
Coll. of Inf. Sci. of Technol., Zhejiang Shuren Univ., Hangzhou, China
Abstract :
This page proposed a new method to synthesize arbitrary length music sequences from a small database of music clips in real-time. The method actually is a patch-base method which is used commonly in texture synthesis of image or motion capture dates. We extracted the music patches automatically from each music clips of the training set. One music patch consisted of three parts: head, body and trail. Both the head part and the trail part were represented by the n-gram pattern feature vectors. To increase the speed of patch matching process, we employed the K-D tree structure to store the feature vectors due to its efficiency and accuracy of multi-dimension searching. Experimental results showed the proposed method is especially useful for interactive and real-time application such as background music generation for games and broadcasting.
Keywords :
audio signal processing; feature extraction; music; pattern matching; real-time systems; tree data structures; K-D tree structure; arbitrary length music sequence; background music generation; image texture synthesis; multidimension searching; music patch extraction; n-gram pattern feature vectors; patch matching process; real-time music synthesis; Databases; Feature extraction; Merging; Nearest neighbor searches; Real time systems; Rhythm; Music analysis; Music retrieval; Music segmentation; Music synthesis; Pattern recognition;
Conference_Titel :
Control Conference (CCC), 2010 29th Chinese
Conference_Location :
Beijing
Print_ISBN :
978-1-4244-6263-6