Dichotic Spectral Integration in Consonant Recognition in Normal Hearing Listeners

The auditory system integrates spectral and temporal cues which are detected and processed by each ear. The dichotic integration ability might be significantly changed when spectral and temporal cues critical for speech perception are modified. In this study, conducted with normal hearing (NH) listeners, we measured spectral integration ranges, which is defined as frequency ranges required for speech recognition in one ear when contrasting (different) frequency ranges are presented in the opposite ear. Dichotic spectral integration ranges were measured under three signal processing conditions. Fourteen NH adult listeners participated in the study. The stimuli included the 14 most frequently used consonants paired with a common vowel /a/. Using the previous data, we intensified only the target frequency and time ranges by 6-dB gain and removed the conflicting frequency and time ranges for each of the 14 consonants. To create contrasting spectral information across ears for all unprocessed and processed consonants, each consonant was low-pass filtered in the left ear, with a fixed cutoff at one of four frequencies: 250 Hz, 500 Hz, 750 Hz, and 1000 Hz. The same consonant was simultaneously presented to the right ear but was high-pass filtered with an initial cutoff of 7000 Hz. Consonant recognition was possible with large spectral information missing (minimum 251 Hz – maximum 4667 Hz) when contrasting spectral information was integrated. Mean dichotic spectral integration ranges (3198 Hz to 4668 Hz) significantly decreased as contrasting frequencies increased. Mean dichotic spectral integration ranges were not significantly affected by the two signal processing conditions (target and target-conflicting) except for the contrasting frequency of 250 Hz. Individual consonant analyses showed that four consonants /ta, da, sa, za/ required less than 50% of frequency information while consonants /fa, va/ required more than 50% of frequency information regardless of the contrasting frequencies. Percentage of the frequency ranges required for consonant recognition was sensitive to the contrasting frequency for four consonants /ga, ma, na, fa/ regardless of the signal processing conditions. The linear regression analyses showed that dichotic spectral integration ranges determined with different contrasting frequencies were linearly related for the target and target-conflicting condition. The results suggest that consonant recognition is possible with large amount of spectral information missing when contrasting spectral information was integrated across ears. Dichotic spectral integration ranges are highly consonant specific and relatively consistent regardless of signal processing on the acoustics useful for consonant recognition. The results suggest that the enhancement of dichotic spectral integration may lead to improved speech perception.