Language-Agnostic Voice Biomarker Platform for Early Detection of Alzheimer’s Disease and Related Dementias (GloVoAD)

BACKGROUND:

Early detection of Alzheimer’s disease and related dementias (ADRD) remains a major clinical challenge, particularly in resource-limited and multilingual settings. Current diagnostic methods, including imaging and cognitive assessments, are often expensive, time-consuming, and typically applied after symptoms have progressed. While voice-based screening has emerged as a promising non-invasive alternative, existing approaches rely heavily on language-specific transcription and are trained on limited, monolingual datasets, restricting their scalability and real-world applicability. There is a critical need for accessible, language-independent screening tools that can detect cognitive decline earlier and across diverse populations.

TECHNOLOGY OVERVIEW:

This University at Buffalo technology introduces GloVoAD, a large-scale multilingual voice dataset combined with a language-agnostic machine learning pipeline for ADRD detection. The platform leverages acoustic vocal biomarkers—such as speech rhythm, pauses, and articulation—rather than linguistic content, enabling detection across languages without transcription. The dataset includes over 2,000 participants spanning multiple diagnoses and languages, with associated demographic and cognitive metadata. The system employs optimized acoustic feature extraction and a cascaded classification architecture that first detects cognitive impairment and then differentiates between ADRD subtypes. This approach enables robust, scalable, and privacy-preserving screening using short voice recordings collected from common devices.

ADVANTAGES:

This technology enables accurate, early detection of cognitive decline using a non-invasive and cost-effective approach. By eliminating reliance on language and transcription, it supports deployment across diverse linguistic populations and reduces bias associated with speech recognition systems. The platform demonstrates strong cross-domain generalizability, maintaining performance across different datasets, languages, and recording conditions. Its use of short audio segments and automated feature extraction enables scalable screening without clinical infrastructure, making it suitable for telehealth, at-home monitoring, and large-scale population screening.

APPLICATIONS:

This technology can be applied in digital health platforms for early dementia screening, remote patient monitoring, and telemedicine. It is suitable for integration into mobile health applications, clinical decision support tools, and insurance or eldercare monitoring systems. It also enables large-scale cognitive health screening in multilingual and resource-limited environments, as well as research applications in neurodegenerative disease detection and progression tracking.

INTELLECTUAL PROPERTY SUMMARY:

Patent 64/010,003 filed March 19, 2026.

STAGE OF DEVELOPMENT:

Technology Readiness Level 6 (TRL 6)

LICENSING STATUS:

Available for licensing or collaboration

ADDITIONAL INFORMATION: