Medical Disclaimer

SomniSense is not a medical device

SomniSense is a wellness and self-monitoring tool. It is not a diagnostic device. It does not diagnose sleep apnea, OSA (obstructive sleep apnea), CSA (central sleep apnea), or any other medical condition.

The Breathing Irregularity Index (BRI) is a product-defined count of candidate acoustic breathing patterns per estimated sleep hour. It is not the clinical Apnea-Hypopnea Index (AHI), and clinical AHI thresholds must not be applied to BRI.

What SomniSense does

• Classifies snore-like audio segments and flags candidate breathing-pattern regions using on-device AI.
• Organizes estimated timestamps, labeled regions, and locally available audio for personal review.
• Compares tagged and untagged nights to show observational associations in your data, not causes.
• Creates structured wellness summaries that you may choose to bring to a clinic visit.

What SomniSense does not do

• Does not diagnose sleep apnea or any other condition.
• Does not replace a licensed clinician's evaluation.
• Does not replace an in-lab polysomnography (PSG) or Type-III home sleep apnea test (HSAT).
• Does not provide medical advice, treatment recommendations, or prescriptions.
• Has not been validated for users under 18.

When to consult a clinician

Consider consulting a qualified clinician if your partner observes breathing pauses, you wake gasping or choking, you experience persistent excessive daytime sleepiness, or sleepiness creates a driving or workplace safety concern. Do not use a BRI value — high or low — to make or delay that decision.

SomniSense observations — tagged nights, locally available audio, and trends — may provide context for that visit, but they are not a substitute for clinical assessment.

Validation context

In published benchmarks using PSG-labeled data from 80 person-nights / 40 participants, 1-second snore-event detection reached 91.67% sensitivity / 89.01% precision (five-seed means), while the compressed breathing model reached 88.49% accuracy and 88.06% F1 on 200-second windows (9,416 parameters, 56.4 KB; 0.064 ms on Apple M2 Neural Engine). These are cohort-level event/window results, not personal-night SRI/BRI accuracy or a diagnosis. Event boundaries, duration, type labels, nightly BRI and personal trends require output-specific validation and do not inherit the window benchmark. Preprints documenting the methodology and the on-device deployment are published openly on Zenodo; U.S. provisional patent applications have been filed and are pending. See /accuracy for full methodology.

Emergency disclaimer

SomniSense is not for medical emergencies. If you or someone else is experiencing a medical emergency, call 911 (United States) or your local emergency services immediately.