Binaural advantage for sound pattern identification

Listeners were trained to identify six patterns of eight sequentially presented 48-ms tone bursts. The variation in frequency forming the patterns was confined to a relatively narrow range around the nominal center frequency, which was either 500, 1000, or 3000 Hz, or was selected randomly on each presentation from a range of 450-3300 Hz. Detection (500 and 3000 Hz only) and identification of the six patterns masked by Gaussian noise was measured in two interaural presentation conditions: masker in-phase and signal in-phase (NoSo), and masker in-phase and signal pi rad out-of-phase (NoS pi). The differences in performance in the two interaural presentation conditions for detection and identification are referred to as "masking-level differences" (MLDs) and "identification-level differences" (IDLDs), respectively. At 500 Hz, MLDs and IDLDs were about 11-13 dB. At 3000 Hz, the MLDs and IDLDs were about 1-3 dB. For the random-center-frequency condition, the slopes of the identification-level functions were much shallower for the NoS pi condition than for the NoSo condition so the binaural advantage was large at low signal-to-noise ratios and declined as signal-to-noise ratio increased. This finding was due to the broad frequency range over which the information was distributed and the decline in the binaural advantage with increasing frequency, a conclusion consistent with that reported for the binaural advantage for speech intelligibility. A second experiment demonstrated that MLDs and IDLDs could be manipulated independently: A 500-Hz tone was added to each element of the 3000-Hz patterns. A large MLD was found--due to detection of the 500-Hz tone--while the identification-level functions were determined solely by the high-frequency information, which produced small IDLDs. Finally, the Gaussian noise masker was replaced by an informational masker comprised of eight randomly chosen eight-tone bursts played simultaneously with the signal-pattern elements which were centered at 1000 Hz. Large amounts of informational masking were found for identification. The slopes of the identification-level functions were much shallower than found for the Gaussian noise masker and a relatively small binaural advantage (about 5 dB) was observed.