水下噪声音色属性回归模型及其在目标识别中的应用

通过对声音的主观评价与客观分析而建立的主观感受数学模型，在许多领域都有重要的应用. 本文采用多元线性回归分析手段对水下噪声音色属性建立回归模型，提取音色特征并改善水下目标的识别效果. 首先，在前期水下噪声音色属性主观评价实验的基础上，将构成音色属性空间的5个成分的评价分值作为回归分析中的因变量，提取大量与听觉感知相关的听觉特征作为自变量；然后,通过相关分析和改进的逐步筛选法，挑选出反映音色属性的“最优”自变量子集；最后，利用向后剔除回归分析和水下目标识别实验,确定适当的音色模型，并通过假设检验证明该线性模型不仅正确有效，而且能改善水下目标识别效果.

Regression model of timbre attribute for underwater noise and its application to target recognition

Timbre attribute is the most important feature to recognize a target. This paper presents a model of timbre features by multiple regression analysis applied in the recognition of underwater noise. At first, timbre attribute as a dependent variable is analyzed by the semantic differential evaluation and principal component analysis. And then an extended stepwise variables selection is proposed to select the optimal set as independent variables from auditory features that have been discussed in previous researches. Finally, the timbre features extracted by the regression model are used to recognize the underwater target. The results show that the extended regression analysis as a statistical method can find the relationship between timbre attribute and the auditory features. And the modeling timbre features calculated by several statistics of the sub-spectral features and the sub-temporal features are more effective than other features.