Sonde is a science and research-driven company
We work with leading organizations to build and validate our voice technology. This research shows that vocal biomarkers provide meaningful health insights to individuals.
Mental Fitness published validation study
The Mental Fitness validation study describes our "Mental Fitness Vocal Biomarker" (MFVB) solution, using vocal analysis to assess mental health via a user-friendly smartphone app. With 104 participants from an outpatient psychiatric population, the study compared the MFVB scores, derived from eight specific vocal features, with traditional mental health assessment tools. Results demonstrated a strong correlation between the MFVB scores and mental health symptom severity, with higher scores indicating better mental fitness. This correlation was especially pronounced among frequent app users, highlighting the benefit of continuous mental health monitoring. The study demonstrates the unique potential of vocal biomarkers as a non-invasive, cost-effective, and privacy-preserving tool for real-time mental health tracking, offering new opportunities for individuals to manage their mental well-being proactively.
Respiratory published validation study
This study by Sonde Health and collaborators demonstrated the ability of a respiratory-responsive vocal biomarker (RRVB) model to differentiate patients with active COVID-19 infection from healthy volunteers. The RRVB model was previously developed on an asthma dataset and also showed strong performance on COPD, interstitial lung disease, and cough, demonstrating its robustness across respiratory conditions. The study suggests that the RRVB model could serve as a prescreening tool for acute respiratory infection and potentially other disease surveillance and monitoring applications in the future.
General Research
A Longitudinal Normative Dataset and Protocol for Speech and Language Biomarker Research
Sonde is committed to advancing the field of vocal biomarker research and has developed the Voiceome protocol, consisting of 12 types of voice tasks, health and demographic factors affecting speech and voice production. In collaboration with Biogen, the Voiceome protocol was used to create a normative dataset with 6,650 participants using Sonde's SurveyLex platform. The protocol itself, code for analysis, and visualization scripts are all publicly available to researchers from SurveyLex and GitHub. This enables easily scalable approaches for decentralized clinical studies in vocal biomarkers and cross-study comparison of outcomes. View PDF
Research supports Sonde’s vocal biomarkers for mental health insights
Medical guidelines, external scientists, and top research institutions have shown that vocal biomarkers are ideally positioned to fill important gaps for early recognition of depression and anxiety risks – An important first step toward engaging and activating individuals to better manage their symptoms and take active steps to stay healthy.
Everyone can benefit from monitoring mental health
In 2016, the US Preventative Services task force issued recommendations that all adults and adolescents be screened for depression. This has become even more important as COVID-19 has disrupted our careers, routines, and families, tripling the share of adults reporting anxiety and depression symptoms.
Today’s methods for recognizing depression risks and symptoms have important limitations
The most accepted tool for assessing depression risk, the Public Health Questionnaire-9 or PHQ-9, relies on honest self-reporting using uncomfortable questions and isn’t optimized for frequent use. In practice, fewer than 5% of people are screened for depression according to established guidelines each year as part of visits to the doctor’s office. This means that the vast majority of depressed individuals routinely go undetected by the healthcare system.
Changes in voice have been recognized as important indicators of depression risk and used in clinical practice since the early 20th century
More recently researchers from prestigious research institutions around the world have built evidence that computational analysis of the speech of individuals can be an effective approach to assessing potential depression and other psychiatric disorders. Sonde’s collaborative research shows smartphones can be effective tools for detecting depression-related speech changes
Mental Health
Vocal Biomarkers for Mental Fitness Scoring and Tracking
Using its technology platform, Sonde is putting the power of vocal biomarkers in your pocket. Anyone can now let their health speak and use their voice for mental health fitness tracking using the Sonde Mental Fitness app. This white paper describes how the app works and what insights users can get.
Novartis collaborated with various groups, including Sonde, to explore novel digital biomarker approaches to detection of depressive symptoms. Sonde's voice technology platform allowed Novartis to build vocal biomarker depression symptom detection models that showed promising performance in identifying these symptoms, consistent with our own research. Notably, among the evaluated technologies, vocal biomarkers are unique in their ability to allow home-based self-assessment providing a quantitative output in a matter of seconds.
Sonde’s speech landmark patterns - acoustic events that correspond to articulatory dynamics within speech - combined with methods developed for use in machine learning approaches to language processing, indicate strong links between human voice and depressive symptom changes.
Speech landmark bigrams for depression detection from naturalistic smartphone speech
Speech landmarks are acoustic events within speech, created by the movement of the articulators (tongue, lips, jaw) and transitions between different speech sounds. Sonde describes how detecting specific combinations and patterns of speech landmarks create strong links between speech and depression.
Differential performance of automatic speech-based depression classification across smartphones
While devices and real-world conditions do impact audio quality and the ability to compute vocal features accurately, this work shows how these challenges can be addressed with the right data and engineering to provide state-of-the-art voice correlations to symptoms of depressions.
Sonde’s research is focused on generalizability and individualization of vocal biomarker measures of mental health. More advanced AI methodologies applied to Sonde’s voice data shows the path toward truly universal speech features and prediction models that can sense changes in depressive symptoms, regardless of the language you speak or the smartphone device model you use.
