Built on peer-reviewed speech biomarker science.

Research from Beth Israel Deaconess Medical Center, UC San Diego School of Medicine, and Harvard Medical School demonstrates that speech-based deep neural networks can model Huntington's disease progression before clinical symptoms appear.

Peer-reviewed research demonstrates that acoustic and linguistic features in the voice can detect manifest Huntington's disease with clinical accuracy.

Clinical research validates acoustic and linguistic voice analysis as a reliable indicator of Huntington's disease progression.

Pringsheim et al. (Movement Disorders 2012) meta-analyzed 13 prevalence studies and found Huntington's disease prevalence of 5.70 per 100,000 in Europe, North America, and Australia, compared with 0.40 per 100,000 in Asia.

A 2002 population-based community validation study (Schrag, Ben-Shlomo, Quinn, J Neurol Neurosurg Psychiatry) found that approximately 20% of patients with Parkinson's disease already in medical attention had not been diagnosed as such, and an additional 15% of patients carrying a PD diagnosis did not meet strict clinical criteria.

Brueckner et al. (2025), in collaboration with Beth Israel Deaconess Medical Center (joint with Harvard Medical School), Northeastern University, and Boston Medical Center, report AUC 0.97 (Sensitivity 0.98, Specificity 0.96, UAR 0.97) for Parkinson's disease detection from natural conversational speech, using features from the HuBERT Large ll60k speech foundation model with a Random Forest classifier. EMBS-BHI 2025 conference proceedings.

Kiyoshige et al. (2025) published in The Lancet Regional Health — Western Pacific report that the inclusion of voice biomarkers significantly improved cognitive-impairment detection AUC from 0.80 (0.76–0.84) to 0.88 (0.84–0.91), and from 0.78 (0.73–0.82) to 0.89 (0.86–0.92), in a cross-sectional study of 1,461 community-dwelling Japanese adults of which 526 (36.0%) had cognitive impairment. DOI: 10.1016/j.lanwpc.2025.101598.

Peer-reviewed research demonstrates that spontaneous conversational speech contains detectable biomarkers for Mild Cognitive Impairment — enabling screening without scripted prompts or clinical interviews.

Clinical research establishes voice analysis as a validated approach to early MCI detection — addressing a condition missed by primary care physicians in 92% of cases.

A real-world health innovation study demonstrates that voice biomarker technology provides objective data for assessing cognitive wellness — improving patient outcomes in community health settings.

A 2011 meta-analysis of primary-care physician accuracy (Mitchell, Meader, Pentzek, Acta Psychiatrica Scandinavica) found GPs recognized 44.7% of MCI cases by clinical judgment and documented the recognition in medical records only 10.9% of the time — about half of MCI cases were missed and the majority of recognized cases went unrecorded.

Mattke et al. observational analysis of the full Medicare population 2015–2019 (Alzheimer's Research & Therapy 2023) found that 7.4 of 8 million (92%) expected MCI cases remained undiagnosed, with substantial racial disparities in dementia detection.

Manly et al. (JAMA Neurology 2022) used the 2016 HRS Harmonized Cognitive Assessment Protocol to estimate that 22% of US adults age 65 and older have mild cognitive impairment, providing a nationally representative prevalence baseline.

With an estimated 46.7 million Americans aged 65 and older living with Alzheimer's or related dementia, peer-reviewed research validates vocal biomarker models for early Alzheimer's detection — identifying cases before clinical symptoms become apparent.

Research demonstrates that Alzheimer's disease indicators are detectable through vocal biomarker analysis in conversational settings — including telephone interactions — making screening possible in any patient encounter.

The Framingham Heart Study, analyzing over 4,000 voice recordings paired with MRI-derived brain data, found that vocal markers including jitter, articulation rate, and lexical diversity are significantly associated with structural changes in memory-related brain regions.

A 2009 Lancet meta-analysis of 50,371 patients across 41 studies (Mitchell, Vaze, Rao) found that unassisted primary-care diagnostic sensitivity for depression was 50.1% — about half of cases were missed at the encounter level.

A January 2026 clinical white paper demonstrates that machine learning models trained on spontaneous speech can serve as an effective first step in identifying individuals at risk for depression and anxiety — enabling earlier intervention at scale.

Peer-reviewed research validates speech-based depression severity detection — enabling nuanced assessment beyond binary present/absent screening.

Clinical research demonstrates reliable audio-based detection of both anxiety and depression through vocal biomarker analysis — validating voice as a dual-condition screening modality.

Research confirms that anxiety and depression are detectable from standard telephone conversations — without specialized equipment, clinical settings, or patient prompting.

A 2015 meta-analysis of 34,902 patients across 24 studies (Olariu et al., Depression and Anxiety) found pooled primary-care diagnostic sensitivity for anxiety disorders was 44.5% — more than half of cases were missed at the encounter level.

Research demonstrates that a single conversational question contains sufficient vocal data to estimate anxiety levels, sleep quality, and mood states through computational voice analysis.

A 2005 study of PTSD prevalence in Veterans Affairs primary care (Magruder et al., General Hospital Psychiatry) found that providers identified only 46.5% of patients diagnosed with PTSD by structured clinical interview — fewer than half. Of identified patients, 47.7% had used mental health specialty services.

A 2003 survey of 600 individuals with bipolar disorder conducted through the National Depressive and Manic-Depressive Association (Hirschfeld, Lewis, Vornik, Journal of Clinical Psychiatry) found that 69% reported being misdiagnosed, most commonly with unipolar depression. Misdiagnosed individuals consulted a mean of 4 physicians prior to correct diagnosis, and over one third waited 10 years or more.

A study of 340 individuals confirms that voice is a reliable indicator of stress — with vocal analysis outperforming self-reported stress measures in clinical accuracy.

Research on voice technology for health monitoring in older adults validates fatigue detection through speech analysis — with direct applications to post-acute and long-term care settings.

Peer-reviewed research demonstrates that fatigue can be extracted as a measurable voice feature — enabling objective clinical assessment without patient self-reporting.

NIH-funded research published in Frontiers in Digital Health — convening experts from MIT, Mayo Clinic, Vanderbilt University Medical Center, and King's College London — confirms voice AI has reached translational readiness for clinical implementation, positioning it as a scalable, inclusive tool for next-generation healthcare.

A systematic scoping review from the University of Málaga, published in Biology (MDPI), confirms that voice production consistently engages the autonomic nervous system — with vocal markers providing measurable signals of stress, cognitive load, emotional state, and subclinical clinical conditions detectable before symptoms appear.

Applied research demonstrates voice biomarker technology reliably identifies stress and vulnerability indicators in conversational settings — with direct applications to clinical care environments.

Powell et al. (Archives of Physical Medicine and Rehabilitation 2008) prospectively assessed 197 ED patients meeting CDC mild TBI criteria and found that 56% did not have a documented mild TBI-related diagnosis in the ED record.

Porter et al. (Journal of General Internal Medicine 2023) estimated that primary care physicians would need 26.7 hours per day to deliver all guideline-recommended preventive, chronic disease, and acute care to a representative 2,500-patient panel — quantifying the structural gap that drives systematic under-delivery of screening elements in primary care.

AUC 0.890 for cognitive decline detection from natural conversation. Verified via confidential strategic clinical-validation partner; partner identity withheld per partnership agreement. Operator-confirmed 2026-05-20.

Internal model performance report (v0.2.0, May 2026) from a confidential strategic clinical-validation partner. Reported AUC values (with sensitivity and specificity): depression 0.874 (Sens 0.722, Spec 0.866); anxiety 0.884 (Sens 0.764, Spec 0.850); PTSD 0.907 (Sens 0.888, Spec 0.745); acute stress and nervousness 0.910 (Sens 0.780, Spec 0.891); psychoactive substance use 0.845 (Sens 0.718, Spec 0.806); fatigue and malaise 0.827 (Sens 0.734, Spec 0.910); ADHD 0.848 (Sens 0.634, Spec 0.868); sleep disorders 0.823 (Sens 0.661, Spec 0.831); bipolar disorder 0.726 (Sens 0.444, Spec 0.860). Partner identity withheld per partnership agreement. Operator-confirmed 2026-05-21.