Keynote-Dependent HMM Based Musical Chord Recognition Method

摘要

Chord sequences and chord boundary is the mid-level performance of the music signal compactness and robustness. Automatic chord recognition is very attractive to researchers in the field of music information retrieval. To improve accuracy of musical chord recognition algorithm, in this study, the importance of keynote was fully considered. According to the music theory, 24 keynotes were defined. For each keynote, a Hidden Markov model was established, which is called keynote-dependent HMM. And then, a recognition algorithm of music chord based on keynote-dependent HMM was proposed. In this algorithm, use MIDI music corpus to train the keynote-dependent HMM, and to improve the recognition rate and facilitate the calculation, a 6-dimensional vector of tonal centroid is used as the feature vector. The experimental results showed that the proposed keynote-dependent HMM had better recognition effect than that of keynote-independent model.