Speech intelligibility is best predicted by intensity, not cochlea-scaled entropy

Andrew J. Oxenham, Jeffrey E. Boucher, Heather A. Kreft

Research output: Contribution to journalArticlepeer-review

9 Scopus citations

Abstract

Cochlea-scaled entropy (CSE) is a measure of spectro-temporal change that has been reported to predict the contribution of speech segments to overall intelligibility. This paper confirms that CSE is highly correlated with intensity, making it impossible to determine empirically whether it is CSE or simply intensity that determines speech importance. A more perceptually relevant version of CSE that uses dB-scaled differences, rather than differences in linear amplitude, failed to predict speech intelligibility. Overall, a parsimonious account of the available data is that the importance of speech segments to overall intelligibility is best predicted by their relative intensity, not by CSE.

Original languageEnglish (US)
Pages (from-to)EL264-EL269
JournalJournal of the Acoustical Society of America
Volume142
Issue number3
DOIs
StatePublished - Sep 1 2017

Bibliographical note

Funding Information:
This research was supported by NIH Grant No. R01 DC012262. We thank Ezleen Mustaffa and Cameron Eller for help with data collection.

Publisher Copyright:
© 2017 Acoustical Society of America.

Fingerprint Dive into the research topics of 'Speech intelligibility is best predicted by intensity, not cochlea-scaled entropy'. Together they form a unique fingerprint.

Cite this