Last year Google enabled spatial audio in head-tracked 360 videos using Ambisonics to binaural decoding on Android mobile devices. There was some early criticism of the 1st order to binaural conversion employed by Google, in terms of the quality of localisation and noticeable frequency response colouration. In this paper, the algorithm used by Google is discussed and the Ambisonics to Binaural conversion using virtual speakers analysed with respect to the resulting inter-aural time, level, and spectrum differences compared to an example HRTF data set. 1st to 35th order Ambisonics using multiple virtual speaker arrays are implemented and analysed with inverse filtering techniques for smoothing the frequency spectrum also discussed demonstrating 8th order decoding correctly reproducing binaural cues up to 4 kHz.
展开▼
