In this paper, we propose a one-to-many neural network (NN)-based model to estimate a personalized head-related transfer function (HRTF). The proposed model comprises a feature representation module and an estimation module. The feature representation module provides a deep feature associated with anthropometric measurement data for a given sound direction. The estimation module is mainly constructed using a bi-directional long short-term memory layer with feature vectors from multiple directions, which results in estimated HRTFs simultaneously for all the multiple directions. The performance of the proposed personalized HRTF estimation method is evaluated using the Center for Image Processing and Integrated Computing (CIPIC) database. Experiments show that the proposed personalized HRTF estimation method reduces root mean square error and log spectral distance by 0.89 and 0.45 dB, respectively, compared to the conventional NN-based method.
(See document for exact affiliation information.)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
