“…The results show that using the first three top predicted HRTF datasets out of 18, 80% of listeners are expected to find one of their favourite HRTF datasets when using the doublehidden-layer neural network. When the number of selected HRTF datasets reaches nine (taking Table 5 as an example, HRTF 16,18,17,7,6,5,11,8, and 4 are selected for Subject 1; taking Table 6 as an example, HRTF 4,16,18,7,10,12,14,6, and 11 are selected for Subject 2), the single-hidden-layer neural network is effective for 100% of population. When using the total error method (Zotkin et al, 2004), 15 HRTF datasets would be needed to be selected to satisfy all of the listeners.…”