Machines Learn to Predict Who Struggles to Hear in a Noisy World

A new systematic review and meta-analysis in the Annals of Biomedical Engineering assessed how well machine learning (and related computational methods) can predict speech-in-noise performance. The authors report an encouraging but cautionary message: models can work, yet not accurately enough for routine clinical use. The paper highlights a major gap between standard audiology tests and real-world difficulty understanding speech in crowded places such as restaurants, train stations, or classrooms. Using PRISMA 2020 guidance, researchers searched PubMed, Scopus, Web of Science, IEEE Xplore, and Google Scholar, screening 6,468 records after deduplication. They included studies using support vector machines, random forests, neural networks, automatic speech recognition, and mechanistic auditory models.