Effect of Gender and Sound Spatialization on Speech Intelligibility in Multiple Speaker Environment

Abstract

In multiple speaker environments such as teleconferences we observe a loss of intelligibility, particularly if the sound is monaural in nature. In this study, we exploit the "Cocktail Party Effect", where a person can isolate one sound above all others using sound localization and gender cues. To improve clarity of speech, each speaker is assigned a direction using Head Related Transfer Functions (HRTFs) which creates an auditory map of multiple conversations. A mixture of male and female voices is used to improve comprehension.

We see 6% improvement in cognition while using a male voice in a female dominated environment and 16% improvement in the reverse case. An improvement of 41% is observed while using sound localization with varying elevations. Finally, the improvement in cognition jumps to 71% when both elevations and azimuths are varied. Compared to our previous study, where only azimuths were used, we observe that combining both the azimuths and elevations gives us better results (57% vs. 71%).