Analyzing Speech to Diagnose Disease

A new technology that uses digital voice biomarkers to assess neurological health could someday be used to identify a variety of conditions, including Alzheimer’s, anxiety, fatigue, depression and PTSD. The patented system combines automatic speech recognition, natural language processing and deep learning to provide an analysis of voice samples in three to five minutes.

Recent research suggests that many neurological conditions can have subtle effects on speech and language that are not distinguishable by brain scans, cognitive tests and observation. However, they can be identified by machine learning. Deep learning algorithms may even provide a means for stratifying patients along a cognitive impairment continuum to determine which individuals are likely to progress to Alzheimer’s disease.

The new approach would enable more patients to be screened and assessed by bringing diagnostic capability to a wearable device. Remote, real-time monitoring would also allow clinicians to track changes over time more easily than traditional laboratory tests. Although the system analyzes elements of speech that are not language-dependent, versions of the technology have been developed for use in other countries, including China and Japan, that are “trained” on native language data samples.