一种混合优化的匹配追踪生态声音识别方法

针对生态自然环境中噪声对声音识别产生干扰的问题,提出利用混合优化的匹配追踪(MP)进行生态声音识别的方法.首先,使用萤火虫算法(GSO)和粒子群算法(PSO)对匹配追踪算法进行混合优化,加快匹配追踪有限次稀疏分解的速度并重构声音信号,保留高相关成分,滤除低相关噪声;其次,根据所选最优原子的时频信息结合MFCCs提取复合抗噪特征;最后,结合支持向量机(SVM)对40种生态声音在不同背景噪声与信噪比的情境下进行分类与识别.实验表明,优化后的匹配追踪算法去噪性能优于谱减法和小波去噪法.与常用的MFCCs方法相比,本方法对生态声音在不同信噪比下的识别性能有不同程度的改善,并且具有较好抗噪性.

Ecological sounds recognition based on hybrid optimized matching pursuit

The paper proposes an anti-noise ecological sounds identification system by using hybrid optimized Matching Pursuit (MP) method. Firstly, using the MP to decompose the sound signal sparsely, reconstruct its high correlation structure and reduce the low correlation noise. Hereinto, Glowworm Swarm Optimization (GSO) and Particle Swarm Optimization (PSO) are employed to speed up the process of MP decomposition. Then, anti-noise composite feature sets are extracted according to the time-frequency information of optimal atoms and the MFCCs. Finally, through the Support Vector Machine (SVM) classifier, 40 classes of ecological sounds are tested for the comparison experiments in different environments under different SNRs. Compared with spectral Subtraction and wavelet de-noising, the MP owns the best performance for de-noising. The experimental results show that this approach outperforms traditional method of MFCCs, as the average identification accuracy and robustness for ecological sounds are improved to a different degree.