The independent contribution of glimpse properties to speech recognition in speech-modulated noise
Bobby E. Gibbs, Daniel Fogerty · The Journal of the Acoustical Society of America · 2017
During fluctuating noise, temporal speech fragments (i.e., glimpses) at sufficient signal-to-noise ratios (SNRs) contribute to speech recognition. Different acoustic properties of these glimpses, related to rate and proportion, reflect the temporal distribution of available speech cues. Glimpse metrics may be used to define different aspects of this temporal distribution. However, these measures are often highly correlated, limiting interpretation of how different glimpse properties contribute independently to speech recognition. The present study investigates how the local SNR cutoff (LC) influences the correlation between glimpse metrics and affects related associations with speech recognition. Speech recognition was assessed in the presence of speech-modulated noise that was temporally manipulated through time compression and presented at different SNRs. Optimization analyses identified LCs that yielded glimpse metrics that were most correlated with the perceptual data. Stimulus manipulations of the noise modulation spectrum and noise level were most related to changes in the glimpse rate and proportion of glimpsed speech, respectively. At an LC of -2 dB, the correlation between these glimpse parameters was minimized while the combined association with speech recognition performance was maximized. Results suggest two glimpse mechanisms, with the relative importance of either mechanism determined by the acoustic noise properties and the analyzed LC.