A query-by-humming system based on marsyas framework and GPU acceleration algorithms

Query-by-humming(QBH) is a content-based music retrieval method and represents the trend of the next-generation search engine. Existing QBH systems still have the problems that most of the retrieval algorithms are of slower searching speeds than the traditional query-by-text ones; lacks a practical QBH system based on the Masyas framework and the GPU acceleration algorithms. This paper introduces a query-by-humming system which is based on the Marsyas framework and GPU acceleration algorithms. The system design, the humming retrieval process, the humming signal feature extraction, the melody libraries establishment and corresponding retrieval algorithms, the GPU parallel speed-up algorithms, and the experiments are presented in detail. The experimental results verify the functionality and good performance of the proposed methods by the speedup of more than 10X with our GPU acceleration algorithms.