Effect of masker modulation depth on speech masking release

Consonant identification was measured for normal-hearing listeners using nonsense Vowel-Consonant-Vowel (VCV) stimuli embedded in a steady-state or fluctuating noise masker and presented at a fixed, global signal-to-noise ratio (SNR) yielding 50% correct identification for steady noise. Fluctuations in the masker were obtained by applying sinusoidal amplitude modulation to the noise. VCVs and noise were either left intact (unprocessed) or degraded by removing temporal fine structure (TFS) cues within 32 frequency bands with bandwidths chosen to match psychophysical estimates of auditory filter bandwidth. For unprocessed stimuli, masking release (MR) - that is better identification in fluctuating than in steady noise - was observed at both masker frequencies tested (8 and 32 Hz). MR increased monotonically and similarly with the modulation depth m of the noise masker (from m=12.5% to 100%) at both masker modulation frequencies. The effect of masker modulation depth on release was different for reception of place of articulation and reception of voicing and manner. The effect of masker modulation depth on release was also significantly affected when TFS cues were removed. These data provide evidence that listeners "glimpse" into noise valleys where maximum SNR may be as low as -5 dB (m=12.5%), and suggest a specific contribution of TFS cues to the glimpsing process.