The field of speech and language technologies is rapidly advancing, with a growing focus on applications in healthcare. Recent developments have shown significant promise in using speech and language models to detect and diagnose neurological conditions such as Alzheimer's disease and related dementias. Acoustic-based approaches, particularly those leveraging automatic speech recognition (ASR) and foundation models, have demonstrated strong potential for scalable and non-invasive detection. Additionally, there is a trend towards developing more diagnostically specific speech perception tests, such as phonetically balanced minimal-pair speech tests, which can better capture perceptual deficits in individuals with hearing loss. Innovative methods, including confidence-based self-training and synthetic data generation, are also being explored to improve the robustness and accuracy of speech reconstruction models. Notable papers include:
- A study that benchmarked foundation speech and language models for Alzheimer's disease detection, achieving high accuracy and AUC scores.
- A paper that introduced a novel methodology for designing and validating a phonetically balanced speech test, using a computational pipeline and ASR system to simulate perceptual effects of sensorineural hearing loss.