TY - GEN
T1 - Correlation Based Glimpse Proportion Index
AU - Alghamdi, Ahmed
AU - Moen, Leonard
AU - Chan, Wai Yip
AU - Fogerty, Daniel
AU - Jensen, Jesper
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - The glimpse proportion (GP) index is an objective intelligibility measure (OIM) based on the glimpse model of speech perception in noise. GP uses local SNR as a criterion to identify time-frequency (TF) regions, or glimpses, that are dominated by speech. Although GP has demonstrated high performance in predicting intelligibility in the presence of stationary and fluctuating noise, its application is limited to additive noise conditions. To address this drawback, we propose a correlation based GP (CGP) index that operates in the TF domain similar to GP but can be applied to a wider range of conditions. The proposed measure is optimized and evaluated using 16 subjective datasets involving speech corrupted by modulated noise, nonlinear processing, and reverberation. The results show that CGP has consistent high performance across all degradation conditions and, on average, outperforms several baseline OIMs. Additionally, CGP has low complexity and takes substantially less time to execute compared to baseline OIMs.
AB - The glimpse proportion (GP) index is an objective intelligibility measure (OIM) based on the glimpse model of speech perception in noise. GP uses local SNR as a criterion to identify time-frequency (TF) regions, or glimpses, that are dominated by speech. Although GP has demonstrated high performance in predicting intelligibility in the presence of stationary and fluctuating noise, its application is limited to additive noise conditions. To address this drawback, we propose a correlation based GP (CGP) index that operates in the TF domain similar to GP but can be applied to a wider range of conditions. The proposed measure is optimized and evaluated using 16 subjective datasets involving speech corrupted by modulated noise, nonlinear processing, and reverberation. The results show that CGP has consistent high performance across all degradation conditions and, on average, outperforms several baseline OIMs. Additionally, CGP has low complexity and takes substantially less time to execute compared to baseline OIMs.
KW - glimpse proportion
KW - speech intelligibility
UR - http://www.scopus.com/inward/record.url?scp=85173049141&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85173049141&partnerID=8YFLogxK
U2 - 10.1109/WASPAA58266.2023.10248110
DO - 10.1109/WASPAA58266.2023.10248110
M3 - Conference contribution
AN - SCOPUS:85173049141
T3 - IEEE Workshop on Applications of Signal Processing to Audio and Acoustics
BT - Proceedings of the 2023 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, WASPAA 2023
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2023 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, WASPAA 2023
Y2 - 22 October 2023 through 25 October 2023
ER -