Optimizing the spatial configuration of a seven-talker speech display

Abstract

Although there is substantial evidence that performance in multitalker listening tasks can be improved by spatially separating the apparent locations of the competing talkers, very little effort has been made to determine the best locations and presentation levels for the talkers in a multichannel speech display. In this experiment, a call sign based color and number identification task was used to evaluate the effectiveness of three different spatial configurations and two different level normalization schemes in a seven-channel binaural speech display. When only two spatially adjacent channels of the seven-channel system were active, overall performance was substantially better with a geometrically spaced spatial configuration (with far-field talkers at −90°, −30°, −10°, 0°, +10°, +30°, and +90° azimuth) or a hybrid near-far configuration (with far-field talkers at −90°, −30°, 0°, +30°, and +90° azimuth and near-field talkers at ±90°) than with a more conventional linearly spaced configuration (with far-field talkers at −90°, −60°, −30°, 0°, +30°, +60°, and +90° azimuth). When all seven channels were active, performance was generally better with a “better-ear” normalization scheme that equalized the levels of the talkers in the more intense ear than with a default normalization scheme that equalized the levels of the talkers at the center of the head. The best overall performance in the seven-talker task occurred when the hybrid near-far spatial configuration was combined with the better-ear normalization scheme. This combination resulted in a 20% increase in the number of correct identifications relative to the baseline condition with linearly spaced talker locations and no level normalization. Although this is a relatively modest improvement, it should be noted that it could be achieved at little or no cost simply by reconfiguring the HRTFs used in a multitalker speech display.