基于生理参数与角度的个性化HRTF深度学习重建方法

Personalized HRTF Deep Learning Reconstruction Method based on Physiological Parameters and Angles

本文设计实现了一个深度神经网络模型,根据人体生理参数及角度信息重建个性化头相关传递函数(Head Related Transfer Function,HRTF),仅需一次训练即可得到全部方向的预测HRTFs。网络模型由将人体测量参数作为输入的深度神经网络、将角度信息作为输入的展开层以及将前两者的输出作为输入的深度神经网络组成。最后对所提出方法的整体性能进行了客观评价。

This paper designed and implemented a deep neural network model to reconstruct a personalized Head Related Transfer Function(HRTF) based on anthropometric parameters and angles information. And the predicted HRTFs in all directions can be obtained with only one training. The network model consists of a deep neural network that takes anthropometric parameters as input,a flatten layer that takes angles as input, and a deep neural network that takes the output of the first two as input. Finally, the overall performance of the proposed method is objectively evaluated.