Skip to content

Biomarkers

Sentinel extracts speech biomarkers from voice audio through specialized models. Each model targets different aspects of mental and physical state, with varying levels of clinical validation.

Biomarker Models

Apollo — Clinical Mental Health

Production Clinical Grade

Domains: Clinical, Mental Health, Employee Wellness, Education, Coaching

Disorder probabilities and symptom-level severities, validated against clinical gold standards. Published in Nature Portfolio.

Disorder Probabilities:

Biomarker Description Range Use Cases
depression_probability Likelihood of depression 0-1 Mental health screening, employee wellness, student support
anxiety_probability Likelihood of anxiety disorder 0-1 Test prep, public speaking, sales calls, interviews

Depression Indicators (PHQ-9 aligned): Clinical Grade

Biomarker Description Range Beyond Clinical Use
anhedonia Loss of interest/pleasure 0-1 Employee engagement, coaching motivation
low_mood Depressed mood 0-1 Wellness monitoring, student mental health
sleep_issues Sleep disturbance 0-1 Fatigue risk, shift worker wellness
low_energy Fatigue/loss of energy 0-1 Workplace safety, sports performance
appetite Appetite changes 0-1 Health screening, wellness programs
worthlessness Feelings of worthlessness 0-1 Coaching, leadership development
concentration Difficulty concentrating 0-1 Education, workplace productivity
psychomotor Psychomotor changes 0-1 Clinical assessment

Anxiety Indicators (GAD-7 aligned): Clinical Grade

Biomarker Description Range Beyond Clinical Use
nervousness Feeling nervous/anxious 0-1 Presentations, interviews, sales calls
uncontrollable_worry Can't stop worrying 0-1 Stress monitoring, coaching
excessive_worry Worrying too much 0-1 Employee wellness, education
trouble_relaxing Difficulty relaxing 0-1 Burnout detection, wellness
restlessness Restless/on edge 0-1 Meeting engagement, education
irritability Easily annoyed 0-1 Contact centers, customer experience, escalation prediction
dread Sense of impending doom 0-1 Crisis detection, mental health

Helios — Wellness Indicators

Production

Domains: Wellness, Workplace, Contact Centers, Education, Coaching, HR

General wellness scores derived from voice characteristics. Uses similar technology to Apollo. Research published at Interspeech 2025.

Biomarker Description Range Use Cases
distress Overall psychological distress 0-1 Crisis detection, customer experience, employee wellness
stress Acute stress level 0-1 Contact centers, workplace, coaching, sports performance
burnout Chronic exhaustion indicators 0-1 HR/retention, education, healthcare workers
fatigue Physical and mental tiredness 0-1 Shift work safety, transportation, healthcare
low_self_esteem Self-worth indicators 0-1 Coaching, education, leadership development

Psyche — Real-Time Affect

Beta

Domains: All (real-time emotion tracking)

Moment-to-moment emotional state from voice. Updates every ~5 seconds, providing real-time signal for live applications.

Biomarker Description Range Use Cases
neutral Flat/neutral affect 0-1 Baseline reference
happy Joy/happiness 0-1 Engagement, satisfaction, coaching success
sad Sadness 0-1 Customer experience, mental health
angry Anger/frustration 0-1 Contact center escalation, conflict detection
fearful Fear/anxiety in voice 0-1 Education anxiety, crisis detection
disgusted Disgust 0-1 Content reaction
surprised Surprise 0-1 Context-dependent

Focus — Attention Tracking

Coming Soon

Domains: Education, Workplace, ADHD Support, Gaming

Attention and focus metrics calibrated for ADHD-aware applications.

Biomarker Description Range Use Cases
focus Overall focus level 0-1 Education, workplace productivity
attention Sustained attention 0-1 Learning, meeting engagement
distractibility Distraction indicator 0-1 ADHD support, driving safety

Diabetes Type 2 Screening

Coming Soon

Domains: Healthcare, Telehealth, Wellness Programs

Voice-based screening indicator for Type 2 diabetes risk.


COPD Screening

Coming Soon

Domains: Healthcare, Telehealth, Respiratory Health

Voice-based screening indicator for COPD.


Configuration

Specify which biomarker models to enable:

sentinel = SentinelClient(
    # ... other config
    biomarkers=["helios", "apollo", "psyche"],
)

Processing Requirements

Different biomarkers require different amounts of speech data. Psyche provides results fastest (~5s), followed by Helios (~10s), with Apollo requiring more speech (~20s) for accurate predictions. These thresholds may vary based on your policy configuration.

Progress Updates

Track biomarker collection progress:

@sentinel.on_progress
def handle_progress(result):
    for name, status in result["biomarkers"].items():
        collected = status["speech_seconds"]
        required = status["trigger_seconds"]
        pct = (collected / required) * 100
        print(f"{name}: {pct:.0f}%")

Validation

Apollo biomarkers are clinical-grade, validated against gold-standard clinical assessments. See the Nature Portfolio publication for validation methodology and results.

Helios wellness biomarkers use similar underlying technology. See the Interspeech 2025 paper for research details.

Clinical-grade means rigorous validation—but the biomarkers are useful far beyond clinical settings: employee wellness, education, customer experience, and anywhere voice reveals what words don't say.