AI-powered automated hearing test okayed by scientists

An error occurred while preparing your download

AI-powered automated hearing test okayed by scientists

An AI-powered hearing test is reliably able to check your hearing on a computer or smart phone without clinical supervision according to a study by University of Manchester researchers.

The high-tech hearing tests, they say, can efficiently understand human speech from the comfort of your own home, rather than at a hospital clinic, by using AI to screen out background noise.

The researchers developed and tested an AI-powered version of the Digits-in-Noise (DIN) test that combines text-to-speech (TTS) and automatic speech recognition (ASR) technologies.

The result was a fully automated, self-administered hearing test that can be performed without clinical supervision in 10 minutes.

The study, funded by a Medical Research Council's Doctoral Training Partnership grant, could revolutionise the way hearing tests are carried out and is published today in the journal Trends in Hearing.

Lead author Mohsen Fatehifar from The University of Manchester said: "Having tested this technology, we are confident that with the help of AI it is entirely possible to automate a hearing test on a computer or smart phone so it can be done from the comfort of your own home.

"Though we still need more extensive trials and a user-friendly interface, this technology could potentially make a huge difference to patients.

"Specialised equipment in the clinic and the specially trained staff who are needed to use it are not always available to patients who need quick assessment.

"Additionally, people are slow to seek help when experiencing hearing difficulties: there is an estimated delay of 8.9 years between the time hearing aids are needed to the time of their adoption.

"That is why we are excited about the ability of this system to incorporate machine learning into the test procedure to make it less dependent on human supervisors."

Speech-in-noise tests are commonly used to detect hearing problems by assessing how well someone can understand spoken speech over background noise.

Traditional tests typically rely on pre-recorded human speech and require a clinician to score the responses.

However, the AI-powered version replaces both with computer generated speech and automatic speech recognition, allowing the test to run entirely on its own.

In a group of 31 adults, some with normal hearing and with hearing loss, the AI-powered test was evaluated against two conventional DIN tests.

The researchers assessed both reliability - how consistent results were across multiple runs and validity - how closely results matched a reference test.

Results showed that the AI-powered test gave virtually the same results as the conventional DIN tests.

While there was slightly more variability in some cases - especially in people with a strong accent- the overall reliability and accuracy were the same, demonstrating the addition of AI did not negatively impact test performance.

And by using larger ASR systems, the researchers say the higher accuracy would make the system compatible with stronger accents.

Co-authors Professor Kevin Munro and Michael Stone are from The University of Manchester and supported by the National Institute for Health and Care Research (NIHR) Manchester Biomedical Research Centre.

Professor Munro said: "This study highlights how AI can make hearing tests both reliable and user-friendly, particularly for individuals who may find traditional formats-such as keyboards or touchscreens-challenging to use.

"It also marks an important step toward more personalised and accessible hearing assessments that people can complete independently at home.

"The test software will be freely available, providing a foundation for future developments using more advanced speech technologies."

Professor Stone said: "This research highlights the potential for well-crafted and tested AI to modernise hearing care.

"Our team plans to explore extending this technology to more complex speech tests in future studies."