John F. Culling
MRC Inst. of Hear. Res., University Park, Nottingham NG7 2RD, UK
Meddis and Hewitt [J. Acoust. Soc. Am. 91, 233--245 (1992)] proposed a model of the use of differences in fundamental frequency to segregate competing sound sources. Key features of the model are: (i) a bank of bandpass filters, to simulate peripheral frequency analysis; (ii) autocorrelation, to determine the periodicities present in each channel; (iii) pooling of autocorrelation functions across channels, to determine the dominant periodicity. Evidence of the dominant source is found by grouping channels containing the dominant periodicity. Evidence of the nondominant source is found in the remaining channels. The model predicts that the periodicity of the dominant source, rather than the nondominant source, determines the ease of segregation. To test the prediction, thresholds for the identification of ``target'' vowels were measured in the presence of ``masking'' vowels. Targets and maskers were composed of sinusoids placed either harmonically or randomly in frequency. Masked thresholds were significantly lower when maskers were harmonic rather than random, but the harmonicity of the targets had little effect. Thus, the results are compatible with the predictions of Meddis and Hewitt's model and are incompatible with models which attempt to compute the pitches of both sources prior to segregation.