Automatic development of speech-in-noise hearing tests using machine learning

Understanding speech in noisy environments is a primary challenge for individuals with hearing loss, affecting daily communication and quality of life. Traditional speech-in-noise tests are essential for screening and diagnosing hearing loss but are resource-intensive to develop, making them less accessible in low and middle-income countries. This study introduces an artificial intelligence-based approach to automate the development of these tests. By leveraging text-to-speech and automatic speech recognition (ASR) technologies, the cost, time, and resources required for high-quality speech-in-noise testing could be reduced. The procedure, named "Aladdin" (Automatic LAnguage-independent Development of the digits-in-noise test), creates digits-in-noise (DIN) hearing tests through synthetic speech material and uses ASR-based level corrections to perceptually equalize the digits. Traditional DIN tests were compared with newly developed Dutch and English Aladdin tests in listeners with normal hearing and hearing loss. Aladdin tests showed 84% specificity and 100% sensitivity, similar to the reference DIN tests (87% and 100%). Aladdin provides a universal guideline for developing DIN tests across languages, addressing the challenge of comparing test results across variants. Aladdin's approach represents a significant advancement in test development and offers an efficient enhancement to global screening and treatment for hearing loss.