Your voice carries more information than words. Tiny shifts in pitch, breathiness, and resonance can signal changes deep inside the body — changes that even trained physicians might miss during a routine checkup. A Berlin-based startup called Noah Labs is betting that artificial intelligence can pick up on those signals before symptoms become critical.
In March 2026, the U.S. Food and Drug Administration granted Noah Labs its Breakthrough Device Designation for Vox, a software-based medical device that analyzes five-second daily voice recordings to detect worsening heart failure. The designation puts Vox on a fast track toward regulatory approval and, potentially, into the phones of millions of patients worldwide.
What Voice Biomarkers Actually Measure
Voice biomarkers are measurable acoustic features — jitter, shimmer, harmonic-to-noise ratio, formant frequencies — that correlate with physiological states. When fluid builds up in the lungs, a hallmark of decompensated heart failure, the vocal cords and surrounding tissues change in subtle but consistent ways.
Noah Labs trained its proprietary AI model on more than three million voice samples collected across five multicenter clinical trials. Partners included the Mayo Clinic and the University of California San Francisco (UCSF). The algorithm learned to map specific acoustic signatures to pulmonary congestion and fluid overload — two conditions that, if caught early, can prevent emergency hospitalizations.
The process for patients is simple. Once a day, they open an app and read a short prompt aloud for about five seconds. The recording is processed on-device and in the cloud, and a risk score is generated. If the score crosses a threshold, the patient’s care team receives an alert.
Why This Matters for Heart Failure Patients
Heart failure affects over 64 million people globally, according to the World Heart Federation. In the United States alone, it accounts for roughly 1 million hospitalizations per year, with 30-day readmission rates hovering near 25 percent. Each readmission costs the healthcare system between $10,000 and $15,000 on average.
Current remote monitoring options include implantable sensors like Abbott’s CardioMEMS and wearable patches, but these are invasive, expensive, or require specialized hardware. A voice-based approach needs nothing more than a smartphone — hardware that 85 percent of American adults already own.
Dr. Elena Ricci, a cardiologist at Massachusetts General Hospital who was not involved with Noah Labs, explained the potential impact: “If we can catch fluid retention two or three days earlier than traditional weight monitoring, we reduce emergency visits significantly. Voice is passive, non-invasive, and costs essentially nothing to collect.”
The Clinical Evidence So Far
Noah Labs has published interim results from its PRE-DETECT-HF trial (NCT07443969), a prospective study enrolling patients with New York Heart Association Class II-IV heart failure. Early findings suggest that Vox identified worsening episodes an average of 4.2 days before clinical deterioration, with a sensitivity of 82 percent and specificity of 79 percent.
These numbers are promising but still preliminary. A larger FDA key trial is expected to begin in mid-2026, and the company anticipates EU approval around the same timeframe, supported by its Horizon Europe funding under the CORDIS framework.
Critics point out that voice biomarker research is still young. Background noise, emotional state, time of day, and even seasonal allergies can affect vocal quality. Noah Labs says its model accounts for these confounders through baseline calibration — the AI first learns a patient’s “normal” voice over a two-week onboarding period before flagging deviations.
Beyond Heart Failure: The Broader Voice AI Movement
Noah Labs is not operating in isolation. The voice biomarker field has attracted serious investment and research attention over the past two years:
- Sonde Health has developed vocal biomarkers for respiratory conditions and mental health screening, partnering with insurance providers to integrate voice checks into telehealth visits.
- Colibri Health focuses on Parkinson’s disease detection through speech analysis, with clinical trials underway at Johns Hopkins.
- Kintsugi uses brief voice samples to screen for depression and anxiety, and its API is already embedded in several employee wellness platforms.
- Researchers at MIT published a 2025 paper showing that vocal features could predict COVID-19 infection with 75 percent accuracy, even in asymptomatic individuals.
The common thread is that voice is a rich, easily collected biomarker. Unlike blood draws or imaging, it requires no appointment, no equipment beyond a microphone, and no clinical setting. This makes it especially valuable for continuous monitoring — tracking a patient’s trajectory over weeks and months rather than capturing a single snapshot during an office visit.
How the Technology Works Under the Hood
At a technical level, Vox uses a pipeline that combines traditional signal processing with deep learning. The audio input goes through several stages:
- Preprocessing: Noise reduction, silence trimming, and normalization to standardize input quality across different devices and environments.
- Feature extraction: The system computes over 200 acoustic features, including Mel-frequency cepstral coefficients (MFCCs), pitch contour statistics, and spectral entropy.
- Classification: A transformer-based neural network processes the feature vectors alongside patient metadata (age, sex, baseline readings) to produce a congestion risk score between 0 and 100.
- Longitudinal tracking: Rather than evaluating each recording in isolation, the model considers trends over the past 7 to 14 days, making it more strong to one-off fluctuations.
The model runs inference both on-device (for immediate feedback) and in the cloud (for the full longitudinal analysis that gets sent to clinicians). Noah Labs says on-device processing takes under 500 milliseconds on a modern smartphone.
Regulatory and Privacy Considerations
Voice data is inherently personal. A recording can identify a speaker, reveal emotional states, and now — potentially — expose medical conditions. Noah Labs stores recordings in encrypted, HIPAA-compliant cloud infrastructure and says raw audio is deleted after feature extraction. Only the computed acoustic features and risk scores are retained long-term.
The FDA’s Breakthrough Device Designation does not guarantee approval, but it does provide several advantages: more frequent interactions with FDA staff during development, priority review of submissions, and the possibility of a less burdensome premarket pathway. For a software-as-a-medical-device (SaMD) product, this can shave months off the approval timeline.
In Europe, Vox is pursuing CE marking under the Medical Device Regulation (MDR), a process that Noah Labs says is running in parallel with FDA activities.
What Comes Next
Noah Labs is planning to expand beyond heart failure. The company has hinted at applications in chronic obstructive pulmonary disease (COPD), where fluid dynamics similarly affect vocal output, and in post-surgical recovery monitoring, where daily voice checks could replace some in-person follow-ups.
The business model appears to center on partnerships with health systems and insurers rather than direct-to-consumer sales. If Vox can demonstrate that it reduces hospitalizations even modestly — say, by 15 to 20 percent — the cost savings for a large hospital network would far exceed the licensing fees.
For patients, the appeal is obvious. Instead of stepping on a scale every morning and hoping their weight catches a problem in time, they talk into their phone for five seconds. The AI does the rest.
Voice biomarker technology is not a replacement for echocardiograms, blood tests, or clinical judgment. But as a screening and early-warning layer, it fills a gap that current tools leave wide open. The next 12 months will determine whether Vox — and the broader field — can turn promising research into routine clinical care.
Related reading on IssueByte:
📚 You May Also Like
Gyucheol Choi
Senior Data Scientist · ImpriMed Korea
AI researcher specializing in precision medicine and drug sensitivity analysis. Covers AI tools, healthcare AI, and emerging tech with hands-on expertise.
Leave a Reply