基于模糊综合和最优小波包分解的信号多类分类

Audio Multi-Classification Based on Fuzzy Integration and Optimal Wavelet Packet Decomposition

为了对音频信号进行有效地分类,提出了基于模糊综合和最优小波包分解的信号多类分类算法。首先,对音频信号进行窗化处理;其次,基于模糊集对信号进行最优小波包分解,并用最优小波包和信号感知特性来提取音频信号特征,在每一个小波子空间用支持向量机对信号进行多类分类;最后,用模糊积分将分类结果进行综合,得出最终类。试验采用不同的核函数和算法参数验证了本文算法的效果,结果表明本算法速度较快、精确度高。

Signal classification is more important in signal process. To perform the effective audio signal classification, an algorithm of the signal feature extraction and the classification is presented based on fuzzy the integration and the optimal wavelet packet. Firstly, window audio signals are proceeded. Secondly, the optimal WP （Wavelet packet） decomposition is used for signals based on the fuzzy set. And the feature of input audio signal is extracted using the optimal WP and perceptual characteristics. Thirdly, this algorithm employs SVM （Support vector machines ） to multi-classify signals in each wavelet subspaces. Finally classification results are integrated by using fuzzy integral. Experiments using different kernel functions and algorithm parameters prove the effect of the algorithm. Results indicate that the proposed algorithm has characteristics of faster speed and higher accuracy.