Predicting Speech Perception in Older Listeners with Sensorineural Hearing Loss Using Automatic Speech Recognition.
Lionel Fontan, Tom Cretin-Maitenaz, Christian Füllgrabe
PMID 32233834WHAT IT FOUND
Automatic speech recognition predicts nonsense-syllable scores well but fails on words and sentences, underestimating human performance because it cannot use linguistic context.
Large prediction errors mean this tool cannot replace clinical speech testing for meaningful speech.
Key findings
01Machine predictions for nonsense logatoms closely matched human scores, but the system dramatically underestimated human performance for words and sentences.
02Prediction errors were very large across all materials, meaning the system could only predict trends rather than individual scores.
03Adding cognitive screening scores improved prediction accuracy only for sentences, not for nonsense syllables or isolated words.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
Testing was conducted in quiet at a lower-than-conversational volume, which does not reflect real-world listening challenges. The hearing loss simulation relied on categorical severity levels, which may have over- or under-estimated effects for individual patients. Cognitive screening was limited to a brief standard test that lacks sensitivity for subtle differences in normal-range cognition.
Declared interests
Funded by the Institute of Advanced Studies at Loughborough University.
The easy way to misread this
Do not assume that strong correlations mean the system can accurately predict an individual patient's speech scores. The large errors and consistent underestimation of word and sentence performance show it cannot replace clinical assessment.