Music Genre Classification Based on Local Feature Selection Using a Self-Adaptive Harmony Search Algorithm

• Main Authors: Yu-Siou Li, 李育修
• Other Authors: Yin-Fu Huang
• Format: Others
• Language: en_US
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/46564348574192057789

碩士 === 國立雲林科技大學 === 資訊工程研究所 === 99 === In this paper, we proposed an automatic music genre classification system based on the local feature selection strategy using an SAHS (self-adaptive harmony search) algorithm. First, five acoustic characteristics such as intensity, pitch, timbre, tonality, and rhythm are extracted to generate an original feature set. Then, the feature selection model using the SAHS algorithm is employed on each pair of genres and then derives their corresponding local feature set. Finally, each one-against-one SVM classifier is fed with the corresponding local feature set and the majority voting method is used to determine the prediction of each music recording. The experiments on the GTZAN dataset were conducted and the results demonstrate our method is effective enough. In the results, the local selection strategies with the wrapper and filter approaches rank the first and third places in the performances among all existing methods.