In this paper, we propose an approach for global HRTF personalization employing subjects’ anthropometric features using spherical harmonics transform (SHT) and convolutional neural network (CNN). Existing methods employ different models for each elevation, which fails to take advantage of the underlying common features of the full set of HRTF’s. Using the HUTUBS HRTF database as our training set, a SHT was used to produce subjects’ personalized HRTF’s for all spatial directions using a single model. The resulting predicted HRTFs have a log-spectral distortion (LSD) level of 3.81 dB in comparison to the SHT reconstructed HRTFs, and 4.74 dB in comparison to the measured HRTFs. The personalized HRTFs show significant improvement upon the finite element acoustic computations of HRTFs provided in the HUTUBS database.
(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.
