AI‐Powered Speech Analysis: Automating Transcription, Embeddings, and Deep Learning for Early Alzheimer's Detection

We have built an AI-based pipeline to transform spontaneous speech recordings into Large Language Model vector embeddings, and compared performance of different machine learning models for Alzheimer's classification. The project source code is hosted on GitHub as an open source project: https://github.com/dzhou08/embedding_AZ In the future, we will integrate this pipeline into a web application as a free service. We will also look into possibilities of combining speech embeddings with other multimodal biomarker data to further enhance the diagnostic accuracy. With the rapid advance of AI and sensor technologies, multimodal frameworks will likely become standard in dementia care in the near future.