A Music Retrieval Method Based on Hidden Markov Model

How to retrieve music accurately and quickly from a large scale of music database has been a hot topic in multimedia information retrieval. In this paper, we aim to propose a novel music retrieval method with the hidden Markov model, which is regarded as the simplest dynamic Bayesian network. We extract two various forms of audio frame features and audio examples features from the audio signal in terms of the unit length in the process of feature extraction. We construct a unique HMM model for each music clip, and several HMM model are used at the same time. The original music is processed by fundamental frequency tracking and state mapping. Afterwards, music information retrieval results can be obtained based on the trained HMM model. Finally experimental results prove the effectiveness of the proposed algorithm.