This study varied the signal-to-noise ratio (SNR) cutoff criterion for acoustically defining usable perceptual glimpses that contribute to speech intelligibility. Criterion-dependent effects were determined by examining the correlation of three different acoustic glimpse metrics with intelligibility. Glimpse properties change depending on the acoustic interactions between the speech and competing noise. Therefore, these measures were investigated with different rates of competing speech that were varied using time compression or expansion. Finally, effects of temporal modulation masking and spectral segregation were examined by comparison between unprocessed (natural) and vocoded speech. Results revealed a range of SNR cutoffs that were associated with correlations between the different acoustic glimpse metrics and intelligibility. Changing the glimpse criterion strongly influenced the associations between intelligibility and two of the acoustic glimpse metrics for the different masker modulation rates. However, the proportion of target speech above the SNR cutoff was less affected by altering the cutoff criterion. These results suggest that intelligibility models should account for the perceptual contribution of different glimpse metrics or limit glimpse cutoff criteria to an SNR region (1–3 dB based on this data) that captures the perceptual utility of multiple glimpse mechanisms.
Cite as: Gibbs, B., Fogerty, D. (2016) Glimpsing Predictions for Natural and Vocoded Sentence Intelligibility During Modulation Masking: Effect of the Glimpse Cutoff Criterion. Proc. Interspeech 2016, 1677-1681, doi: 10.21437/Interspeech.2016-1587
@inproceedings{gibbs16_interspeech,
author={Bobby Gibbs and Daniel Fogerty},
title={{Glimpsing Predictions for Natural and Vocoded Sentence Intelligibility During Modulation Masking: Effect of the Glimpse Cutoff Criterion}},
year=2016,
booktitle={Proc. Interspeech 2016},
pages={1677--1681},
doi={10.21437/Interspeech.2016-1587}
}