Influence of competing multi-talker babble on frequency-importance functions for speech measured using a correlational approach

In this study, a correlational approach was used to estimate the relative importance of five different frequency bands for the identification of speech in two listening conditions: In the first, the nonsense vowel-consonant-vowel (VCV) signals that the listener had to identify were presented in the absence of other potentially interfering speech signals. In the other condition, the target VCVs were presented in a multi-talker babble background. The signal-to-babble ratio (SBR) was fixed at 5 dB. Fifteen young normal-hearing listeners were tested. Each listener was successively presented with 1500 stimuli in each condition. On each presentation, a randomly selected VCV signal was added to noise in five contiguous frequency bands within a range from about 100 to 7750 Hz. The signal-to-noise ratio (SNR) was roved independently in the different bands over a 24 dB range, the mid-point of which was adjusted so that listeners achieved between 60 and 70% correct in both test conditions. The importance of each frequency band was estimated by computing the point bi-serial correlation coefficient between the successive SNRs in that band and the corresponding binary identification scores (correct/incorrect) across the 1500 trials. The shapes of the importance functions measured in the two conditions were found to be significantly different: the importance of the middle-to-high frequency band was significantly lower in the presence of babble, while the importance of the two lowest bands was significantly higher. Possible reasons for this specific finding and more general considerations regarding the application of the cor-relational method are discussed.