关键词: 特征提取, 音频检索, 熵值法, 矩阵相似度

Abstract: Through analyzing the feature extraction and the matching algorithm, it proposes a completed framework of audio data retrieval system. The system mainly analyzes the feature extraction and the matching of feature. In the part of audio feature extraction, it analyzes the classical MFCC coefficient and proposes a weighted MFCC coefficient based on the entropy value method, which improves the recognition rate of retrieval. In the part of audio matching, it uses the characteristic parameter matrix to represent the property of audio information, and introduces a matching method of matrix similarity, which improves the retrieval efficienty. The experimental results show that the recognition rate of system increases 1.2% and the time decreases 22%, and the performance of system is improved obviously.

Key words: feature extraction, audio retrieval, entropy value method, matrix similarity