GPU (graphics processing unit) based melody matching parallelization method

Abstract

A GPU based melody matching parallelization method comprises four steps as follows: step one, performing calculation preparation work on a CPU (central processing unit); step two, performing variable initialization operation on a GPU before calculation; step three, performing actual DTW (dynamic time warping) calculation on the GPU; and step four, organizing follow-up results on the CPU. According to the method, a computation sequence of accumulated distance matrixes in a DTW algorithm is changed, and data dependence among matrix elements during calculation is eliminated with a diagonal calculation method, so that a single DTW calculation is accelerated by using the GPU; and then, the diagonal calculation method and parallel calculation of multiple DTW calculation tasks are combined, so that multiple DTW calculations can be accelerated by the GPU. According to the method, the programming is simple, the speed-up ratio and the performance cost ratio are high, and the method has a wide application prospect in the technical field of computers.

Description

technical field [0001] The present invention proposes a parallelization method for melody matching based on GPU, and specifically relates to a method for parallelizing the melody matching algorithm in humming retrieval from the data calculation method of the algorithm itself and the execution process of multiple algorithms by using GPU , belonging to the field of computer science and technology. Background technique [0002] Content-Based Audio Information Retrieval (CBAIR) is the future development trend in the field of music retrieval, which is different from the traditional classification and retrieval methods based on text description. Carry out automatic classification and replace manual text description. Query By Humming (QBH) is a main retrieval method in content-based music retrieval. Compared with traditional music retrieval methods, humming retrieval is more in line with Music is a query work that has a highly subjective carrier. The main problem of humming retri...

AI Technical Summary

Problems solved by technology

The main problem of humming retrieval is the lack of user experience. Due to the high complexity of the melody matching algorithm used in humming retrieval, the time required for retrieval is relatively long, and the waiting time for users is correspondingly long

Note that the matrix D can be calculated in the form of column priority, that is, calculating the value of the m-th column element only requires the value of the m-1th column element, that is, if we only need DTW(X,Y)=D(N, M) value, then the required storage space is O(N), similarly, if the row-first method is used for calculation, then the required storage space is O(M), but no matter how it is calculated, the time complexity is always is O(NM)

Images