空间听觉重建中,头相关传输函数(head-related transfer function,HRTF)庞大的数据量是影响虚拟声源合成效率的主要因素之一.为了减少HRTF的数据存储,提出一种局部线性嵌入(locally linear embedding,LLE)空间听觉重建方法.通过LLE对高维HRTF数据进行降维,在低维数据空间提取与方位感知相关的特征,然后利用聚类算法进行分类,得到特征HRTF,而其余非特征HRTF则可以利用特征HRTF通过改进插值算法进行重构.与现有的主成分分析法(principal component analysis,PCA)相比,利用LLE降维后的数据保留了更多的感知信息,利用HRTF数据间的内在关系,对插值后的数据进行修正,可减少重建误差.仿真结果表明,该方法能够有效地减少HRTF的存储数据量,有利于提高虚拟声源的合成效率.
To improve localization accuracy, the spherical microphone arrays are used to capture high-order wavefield in- formation. For the far field sound sources, the array signal model is constructed based on plane wave decomposition. The spatial spectrum function is calculated by minimum variance distortionless response （MVDR） to scan the three-dimensional space. The peak values of the spectrum function correspond to the directions of multiple sound sources. A diagonal loading method is adopted to solve the ill-conditioned cross spectrum matrix of the received signals. The loading level depends on the alleviation of the ill-condition of the matrix and the accuracy of the inverse calculation. Compared with plane wave decomposition method, our proposed localization algorithm can acquire high spatial resolution and better estimation for multiple sound source directions, especially in low signal to noise ratio （SNR）.
采用主成分分析方法提取头相关传输函数(head-ralated transfer function,HRTF)的个性化系数,计算了影响HRTF的人体参数的拉普拉斯得分,并联合Pearson相关系数提取出对HRTF影响显著的关键人体参数;构建了径向基函数(radial basis function,RBF)神经网络,学习关键人体参数到头相关传输函数个性化系数的非线性映射模型,利用简单的人体参数测量估计出待测者的个性化头相关传输函数.通过实验仿真与偏最小二乘回归(partial least squares regression,PLSR)法比较可知。
