Digits-In-Noise Hearing Test Using Text-to-Speech and Automatic Speech Recognition: Proof-of-Concept Study.
Mohsen Fatehifar, Kevin J Munro, Michael A Stone and 3 others
PMID 41032630WHAT IT FOUND
A self-supervised hearing test using AI to generate and recognise speech worked as well as standard tests only for participants with clear accents.
The system failed to reliably score responses for several people, meaning it cannot yet replace clinician-administered testing for diverse populations.
Key findings
01The AI-powered test showed comparable validity to standard tests, but reliability was worse unless participants with high speech recognition error rates were excluded.
02Five participants were excluded from validity analysis and seven from reliability analysis because the automatic speech recognition system made too many errors.
03Synthetic speech stimuli generated by text-to-speech technology had intelligibility and naturalness scores similar to human-recorded stimuli.
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
The study was conducted in a sound-treated booth with calibrated equipment, not in a home environment where the test is intended to be used. Participants with strong accents were excluded if the speech recognition error rate exceeded 40%, meaning the reported success rates do not reflect performance for the full population. The sample size was small (31 participants) and recruited via convenience sampling from a university and audiology centre. The test used only digits, which is a simpler task than sentences or words, potentially overestimating the system's ability to handle more complex speech. The study did not verify the quality of the synthetic speech stimuli before presenting them to participants, which could have introduced errors not caused by the patient.
Declared interests
The authors declared no potential conflicts of interest. The work was supported by the Medical Research Council and the NIHR Manchester Biomedical Research Centre.
The easy way to misread this
Do not interpret the high correlation scores as evidence that AI hearing tests are ready for widespread clinical deployment. The results rely on excluding up to 22% of participants because the software could not accurately recognise their speech, a failure mode that is unacceptable for a screening tool intended to be used by diverse populations without supervision.