An international consortium of researchers has published the first consensus-based framework and standardized taxonomy for vocal biomarkers, addressing a critical barrier to the development of voice-based health technologies. The new guidelines, published in the journal Digital Biomarkers, aim to replace inconsistent terminology with a common scientific language that can support research collaboration, clinical validation, and regulatory approval.
The effort was led by the Department of Precision Health at the Luxembourg Institute of Health (LIH) and the University of South Florida Morsani College of Medicine, through the European eVoiceNet and North American Bridge2AI-Voice networks. Twenty-four experts from Europe and North America participated in a multi-stage consensus process between 2024 and 2025.
Establishing a common language for voice health
As research into voice-derived health indicators has grown, terms like "voice biomarkers," "speech biomarkers," and "vocal biomarkers" have been used interchangeably, despite referring to different physiological and cognitive processes. This lack of clarity has slowed collaboration between clinicians, engineers, data scientists, and regulators. The new framework provides a scientifically grounded vocabulary designed to improve these interactions, a goal that aligns with the cross-disciplinary demands of AI for Science & Research.
The framework clearly distinguishes between a vocal measure and a validated vocal biomarker. A vocal measure is any quantifiable acoustic, linguistic, or respiratory parameter-such as pitch perturbation or pause length. A vocal biomarker is a specific measure or combination of measures that has undergone formal clinical validation to reliably indicate a disease diagnosis or physiological state. The taxonomy also introduces a hierarchical model that spans the different domains involved in voice and speech production.
Tracking disease through voice
Research shows that subtle changes in speech, breathing, and vocal quality can provide clues about conditions such as Parkinson's disease, Alzheimer's, depression, heart failure, and type 2 diabetes. Vocal biomarkers capture information from multiple physiological and cognitive systems simultaneously, making them a rich source of digital health data. This monitoring of health conditions through voice analysis is an active area of development in AI for Healthcare.
"Voice has enormous potential as a source of health information, but the field cannot progress efficiently without a common language," said Dr. Guy Fagherazzi, head of the Department of Precision Health at LIH and chair of eVoiceNet. "By defining what we mean when we talk about voice-based health measures, we are creating the foundations for more robust research, greater transparency and, ultimately, clinically useful technologies that can benefit patients."
"One of the unique strengths of vocal biomarkers is that they capture information from multiple physiological and cognitive systems simultaneously," said Dr. Yael Bensoussan, associate professor of Otolaryngology at USF Health and co-head of the Bridge2AI-Voice consortium. "However, this complexity is also what has made the field difficult to define. This work provides a structure that allows researchers to speak the same scientific language while preserving the richness of the signal."
Why this matters for science and research professionals
The VOCAL framework supplies a shared vocabulary that can streamline study design, data analysis, and cross-disciplinary collaboration. It also lays the groundwork for future validation standards and regulatory guidance, potentially accelerating the translation of voice-based diagnostics from the lab to clinical practice. For researchers working at the intersection of digital health, AI, and medicine, this consensus marks a step toward more reproducible and clinically meaningful tools.
Your membership also unlocks:
