University of Oldenburg, Medical Physics
This study presents a model of the psychoacoustic study of Kopčo et al. [J. Acoust. Soc. Am. 127(3), March 2010] that measured detection and localization performance for a female voice target among four male maskers, distributed in the horizontal plane in a reverberant room. All voices simultaneously uttered one monosyllabic word, leading to a complete temporal overlap of target and masker signals. The target utterance was fixed throughout the experiment, whereas its azimuth was randomized across trials. Several spatial configurations of the random masker utterances were tested in two conditions: Either the masker configuration was fixed throughout a run, or it varied randomly from trial to trial. The proposed model integrates binaural and harmonic cues extracted using existing auditory peripheral models of binaural and harmonicity processing. A template of the time course of the harmonic cues from the target voice alone was used to identify the target in the mixture. Thus, the harmonic cues can be seen as a prior guiding the readout of binaural features related to the target. Additionally, the model integrates cues across frequency bands. Model results show a reliable localization ability of the target voice in the multitalker mixture that is comparable to the psychoacoustic data.
ICS file for iCal / Outlook