Frequently Asked Questions
Partnering with Hyfe for research
Partnering with Hyfe for research
Hyfe is a digital health company that develops privacy-preserving, on-device AI to continuously detect and measure cough in real-world settings. Its technology enables objective, continuous longitudinal cough data for clinical trials, research, therapeutic applications as well as healthcare applications.
Historically cough has been difficult to measure reliably. Traditional methods are subjective, episodic, or burdensome, which limits their scientific utility. As a result, cough has been underrepresented in endpoints, despite its relevance across many diseases. This has slowed down innovation in cough medicine to the point that the last antitussive approved by the FDA was in the 1950s. Over the last 13 years alone at least 15 candidates have either failed or have been rejected by the FDA .
Cough is one of the most common and clinically meaningful symptoms in medicine, yet it has historically been measured poorly or not at all. Traditionally clinical evaluations of cough have been limited to patient recall or short, artificial recordings. This has also led to a stagnation in innovation in cough medicine because it has been very hard to generate clinical evidence that would satisfy a modern regulator. Hyfe addresses this gap by enabling continuous, objective cough measurement in the real world and at scale.
Hyfe performs real-time cough detection using an AI model that runs directly on the device without recording or storing audio. This enables continuous monitoring over weeks or months, preserves privacy, reduces burden, and generates longitudinal data that is not feasible with audio-recording approaches. The recording runs typically on a stylish, discrete and waterproof wearable that is almost identical with an apple watch. Learn more about it here.
Hyfe is widely used by clinical trial teams studying respiratory and cough-related conditions, pharmaceutical companies, CROs, academic researchers and digital health partners. Hyfe’s cough detection models also integrate into third party platforms and devices.
The technology is also available to consumers, in a free application available on iPhone and Android.
Hyfe’s products include:
Hyfe is best described as a digital biomarker and digital endpoint company. Its core asset is validated, fully privacy-preserving continuous cough monitoring AI that transforms ambient sound into clinically meaningful, quantitative cough data.
Hyfe’s Vision is to make continuous cough monitoring a cornerstone of healthcare, enabling earlier diagnoses (including preventing hospitalizations), better and personalized treatments, and improved quality of life.
Cough reflects disease activity, exacerbations, and treatment response in multiple conditions. When measured objectively and longitudinally, it can provide actionable insight that complements or improves upon traditional endpoints. Being able to measure cough reliably in clinical research settings enables pharma to bring to market a new generation of molecules that address cough directly or indirectly.
Continuous cough monitoring endpoints increase data density, reduce reliance on recall, and capture real-world variability, improving statistical power, interpretability, and confidence in outcomes.
Exploratory continuous cough endpoints enable early antitussive signal detection, while secondary and primary objective cough endpoints support market differentiation and enable labeling claims.
Peer-reviewed studies have demonstrated correlations between cough frequency and disease severity, exacerbations, and treatment effects across multiple respiratory conditions. Continuous monitoring has further shown predictable patterns and clinically relevant variability.
By operating passively in the background, requiring no active input, no diaries, and no audio review (no one can listen to patient conversations or access any audio data at any point.), while still generating high-resolution, analyzable data. Hyfe’s CoughMonitor Suite is convenient, discrete and familiar, leading to superior adherence and compliance with no participant burden.
Hyfe is used to generate objective, continuous cough frequency data both as an endpoint and as an operational signal (e.g., screening, adherence, change-from-baseline). It is used extensively across Pharma trials, investigator-initiated as well as sponsor-led studies throughout phases 1 - 4.
Yes. Objective cough frequency is commonly used as a primary endpoint in antitussive development. While historically it has been measured over a defined, limited window (typically 24 hours), new evidence and the emergence of technologies such as Hyfe, allow for much longer - and much more precise - cough monitoring.
When cough is clinically relevant but not the primary therapeutic target, e.g., as a symptom endpoint, a treatment-response marker, or a supportive/ bridging measure alongside PROs, spirometry, imaging, or event-based outcomes.
By enabling objective baseline cough quantification before randomization. This supports eligibility confirmation, exclusion of low-cough participants when appropriate, and enrichment strategies based on measurable symptom burden.
Continuous monitoring increases data density and enables within-subject longitudinal analyses, which can improve sensitivity to change and reduce noise relative to sparse sampling or recall-based measures. This typically improves power for a given N.
Passive measurement reduces behavioral workload, which leads to improved adherence in practice. Retention effects should be treated as empirical and study-dependent rather than assumed.
It reduces reliance on manual cough logs and episodic recordings and provides centralized, structured cough metrics that can be reviewed without audio handling. This significvantly reduces site-level operational complexity.
Yes. Hyfe aligns with remote data acquisition models by collecting cough measures in daily life and transmitting derived data for centralized review. This is consistent with FDA guidance encouraging fit-for-purpose digital health technologies for remote acquisition.
Deployment is largely governed by standard trial steps: protocol fit, privacy/security review, vendor qualification, and integration requirements, rather than physical site hardware. Timelines are study-specific. Once these steps are met, Hyfe can deploy in a matter of days.
Absolutely - Hyfe’s CoughMonitor Suite has been used (and continues to be used) in sponsor-led studies up through Phase 2b.
By enabling a fit-for-purpose validation pathway:
Hyfe has extensive, rigorous, publicly available data on validation work.
Yes, if collected under a protocol that supports data integrity, traceability, and fit-for-purpose validation.
Hyfe can support evidence generation needed for qualification by producing standardized, objective measures and validation packages aligned to a defined Context of Use. Formal qualification typically follows established programs such as FDA’s Drug Development Tool qualification pathways and EMA’s Qualification of Novel Methodologies process
Hyfe’s cough detection tools have been used in dozens of field and clinical studies and have been evaluated in peer-reviewed work including:
Published work includes chronic cough and broader respiratory applications (e.g., COVID-era respiratory disease monitoring), as well as use across conditions such as asthma, COPD, IPF, Bronchiastasis, congestive heart failure, and others.
Validation and evaluation studies include adults with a broad range of cough-associated diagnoses in real-world daily living, as well as participants studied in prospective cohorts using wearable or smartphone based monitoring.
Hyfe has been evaluated both in controlled reference setups (e.g., protocols with ground-truth annotation) and in real-world, free-living monitoring.
Yes. Hyfe’s tools are used by academic groups, and the platform is positioned to support investigator-led studies where cough is an endpoint or an exploratory measure, subject to protocol fit and governance requirements. Go to hyfe’s Research Portal
Cough exhibits significant circadian and day-to-day variability. Single-timepoint measurements fail to capture this structure, while continuous monitoring reveals stable individual patterns and clinically meaningful deviations.
Cough reflects airway irritation, inflammation, and neuro-respiratory signaling. It is present across a wide range of conditions, including respiratory, cardiac, infectious, and fibrotic diseases, making it a broadly informative clinical signal rather than a disease-specific symptom.
Cough frequency has demonstrated relevance in conditions such as chronic cough, asthma, COPD, bronchiectasis, interstitial lung diseases, respiratory infections, heart failure, and long COVID.
Changes in cough frequency often precede or accompany clinical deterioration, decompensation periods, exacerbations, and therapeutic effects. Longitudinal data shows that cough dynamics can reflect both acute events and longer-term disease trajectories.
Unlike patient-reported outcomes, cough frequency is objective and continuously measurable. Unlike spirometry or clinic-based assessments, it captures real-world disease expression outside controlled environments.
Patient-reported measures are affected by recall bias, perception variability, and reporting fatigue. They compress a dynamic behavior into coarse summaries, limiting sensitivity to change.
Yes. There is emerging evidence that shows that deviations from individual baseline cough patterns can precede clinically recognized exacerbations or disease onset.
They are episodic, burdensome, and difficult to scale. Short recordings cannot represent longer-term variability, and raw audio introduces privacy, storage, and review challenges.
It refers to passive, ongoing measurement of cough events across daily life over extended periods - days, weeks, or months - without requiring user interaction or manual data review. Hyfe’s detection models run passively on a discrete, water proof wearable device which enables continuous, real world data over any length of time.
Dense time-series data increases sensitivity to change, enables within-subject analyses, and reduces noise introduced by recall or sparse sampling, often improving power without increasing sample size.
Cough frequency is most powerful when interpreted longitudinally and, where appropriate, alongside other clinical or digital measures. Its strength lies in objectivity and temporal resolution.
Yes. Objective detection enables consistent definitions and measurement across sites and geographies, supporting comparability and reproducibility.
Because measurement technology constrained research questions. Advances in on-device AI and continuous monitoring have removed those constraints, enabling modern cough science.
Hyfe uses machine-learning acoustic models to identify cough events from ambient sound. Detection is performed as a classification task on short audio features, producing time-stamped cough counts and related summary metrics. More here.
No. Hyfe is designed to detect cough events, not to interpret speech content. The detection model runs on device in real time. It does not need to capture or store speech - or any sound for that matter - to perform cough detection.
Hyfe’s AI models are trained on an extensive dataset of real-world coughs. Through machine learning it learns discriminative acoustic signatures of cough - things like temporal envelope, spectral structure, and transient characteristics - that differ from speech, laughter, throat clearing, and common environmental noises. The system applies decision thresholds and post-processing rules to reduce confusion with cough-like sounds.
While Hyfe’s models are fundamentally different from anything else on the market, three differences matter th emost in practice:
On-device inference means the model runs locally on the device (coughWatch, wearable, phone, etc) rather than streaming audio to a server for processing. It matters because it materially improves privacy, reduces bandwidth and infrastructure needs, enables longer monitoring periods, and lowers operational risk in regulated settings.
Hyfe’s models are trained and tuned to operate in uncontrolled environments and use signal conditioning, robustness features, and conservative decision logic. They are trained on vast amounts of highly heterogenous data, including any possible permutation of hardware, software, socio-demographic background and user behaviour. They are trained exclusively on real-world data and are highly optimized to function reliably in the real world.
Hyfe’s detection performance is evaluated against ground truth in validation studies. Accuracy is best summarized in terms of sensitivity and specificity under defined conditions rather than a single universal number, because real-world noise and device placement materially affect results.
Training uses large, labeled datasets that include cough and non-cough acoustic events across varied real-world conditions. Validation is performed by comparing model outputs to independently established ground truth (e.g., human-labeled reference data) and reporting standard performance metrics, with attention to generalization beyond the training set.
Environment and device placement typically have a larger impact than language or accent, because cough acoustics are not language-dependent while noise profiles vary widely. That said, Hyfe evaluates generalization across diverse conditions and monitors for performance drift in new contexts. The Models are trained on data collected in the real world across +160 countries.
By combining (1) robust feature learning, (2) threshold calibration for the intended use case, and (3) temporal consistency checks that filter isolated, cough-like artifacts. The trade-off between false positives and false negatives is managed explicitly, depending on whether the application prioritizes sensitivity (e.g., screening) or specificity (e.g., endpoint integrity).
Hyfe’s core output is cough event detection and frequency over time. Pattern-level insights, such as bursts, time-of-day structure, clustering are feasible from the time series/ frequency. Differentiating clinical “cough types” is possible in principle but depends on the specific labeling scheme, evidence base, and intended decision use and is not something Hyfe is currently pursuing. Learn more about the different acoustic AI frameworks.
Hyfe can run on common consumer devices capable of background audio sensing and on-device ML inference - essentially any device with a basic chipset and a microphone. Hyfe’s CouughMonitor Suites includes the CougWatch, a wearable optimized for cough detection. The models are also natively integrated in medical grade wearables such as the Actigraph LEAP. The models also run reliably in most consumer grade smartphones - see CoughPro.
Yes, as long as the wearable platform has a microphone.
No. It is designed to work with standard microphones and consumer compute. Performance can vary by device quality and placement, but specialized sensors are not required.
Absolutely. The intended operating mode is passive monitoring with minimal/ no user interaction, enabling longitudinal measurement without diaries or active tasks.
Hyfe outputs time-stamped cough metrics that can be aligned with other time-series data (activity, sleep, physiology) for multimodal analyses. Integration is typically done via firmware, but it is also possible to do via shared timestamps, standard exports/APIs, and study-specific analysis plans that define how cough complements other endpoints.
In standard operation, Hyfe’s clinical-trial systems are designed so no raw audio is recorded or sent off-device; audio is processed transiently to make a cough/no-cough decision and then discarded.
In some SDK deployments, optional audio sharing can be enabled only with explicit opt-in consent for defined purposes (e.g., validation or research), with a separate tech stack and separate disclosure.
Hyfe is built around data minimization: it detects cough events on the device and outputs non-identifiable metadata (e.g., timestamps/counts) rather than collecting identifiable content. The system is designed so Hyfe does not receive participant identifiers; study teams manage any link between a participant and a device/study ID or any PHI as per HIPPA guidelines.
Cough detection occurs on-device (on-device inference). Depending on the configuration, derived outputs (e.g., cough timestamps and durations) are transmitted to Hyfe servers for aggregation/ analytics.
Hyfe’s approach aligns with common regulatory expectations by minimizing personal data, limiting identifiability, and applying standard transfer safeguards where relevant (e.g., Standard Contractual Clauses for international transfers). Hyfe is GDPR and HIPAA compliant as well as ISO 27001 compliant.
Cloud audio-recording approaches create persistent raw data that is inherently higher risk (storage, access control, breach impact, review workflows, and IRB scrutiny). Some solutions on the market even rely on human operators who label (and therefore listen to) continuous recordings.
Hyfe’s architecture is designed with privacy as a first principle.. It avoids audio retention entirely and is relying on derived event data instead.
Because raw audio is commonly treated as higher-risk data by participants, IRBs, and sponsors. Minimizing or eliminating retained audio reduces perceived surveillance risk, simplifies governance, and lowers operational overhead, making longer monitoring windows and larger deployments more feasible.
Hyfe uses encryption in transit (TLS) and encryption at rest (AES), with storage on compliant cloud infrastructure and role-based access controls.
For SDK/API analytics, Hyfe also uses encrypted, full-volume encrypted server storage for data sent to the Insights API.
Hyfe adheres to GDPR, HIPAA, and other applicable laws for its clinical-trial platform and structures data flows to support those obligations (minimization, security controls, transfer safeguards). Final compliance in any deployment depends on the sponsor’s study design, roles (controller/processor), and contractual terms.
By providing a system architecture that limits identifiable data, maintains clear data lineage (what is collected, what is not), and applies standard security controls and retention practices, reducing the scope and complexity of privacy and security review.
Hyfe has in place enterprise-grade controls including pseudonymization, role-based access, defined retention, and safeguards for cross-border transfers. Study identifiers are managed by the sponsor/investigator, and Hyfe is designed to operate without access to direct identifiers.
Yes! Cough monitoring plays an important role in therapy when it enables feedback, awareness, and behavior change, or when it supports timely clinical action. The therapeutic effect comes from how the signal is used (feedback loops, coaching, clinical workflows), not from measurement alone. Additionally, cough monitoring is a key component of modern digital therapeutics that focus specifically on refractory chronic cough
Objective feedback reduces reliance on perception and recall. When people can see cough frequency and trends, it can improve self-awareness, support goal setting, and reinforce behavioral strategies. Similar to how step counts or sleep metrics influence health behaviors.
Behavioral cough suppression therapy (BCST) relies on identifying triggers, practicing suppression strategies, and tracking progress. Continuous monitoring can provide objective reinforcement (did cough decrease?), characterize time-of-day and trigger-linked patterns, and help distinguish true improvement from day-to-day variability. See more
Hyfe can provide the measurement layer required for a digital therapeutic: passive cough quantification, longitudinal tracking, and change-from-baseline metrics. A therapeutic product still requires a validated intervention (content, coaching, clinical claims, and regulatory strategy) built on top of that measurement. Hyfe is already working with partners in Pharma to bring a new generation of digital therapeutics to market - i.e. in Japan
Yes. Hyfe’s tech is used daily in research cohorts and care-oriented programs to generate longitudinal cough data in everyday settings, not only under trial protocols. It is even available in consumer-facing, wellness products such as CoughPro
By producing objective, time-stamped cough measures collected passively during daily life. This supports longitudinal analyses (baseline, trajectories, change-from-baseline) and enables linking cough dynamics with exposures, comorbidities, and treatment changes captured elsewhere.
At the population level, cough time series can be analyzed for associations with seasonal patterns, outbreaks, and environmental exposures. There is literature described using longitudinal cough data to surface patterns and potential triggers such as seasonal allergies.
Additionally, continuous cough monitoring can surface environmental factors for indicvidual users - helping define individual triggers for allergies or asthma for example.
Long-term monitoring reveals individual baselines and day-to-day variability, making it possible to detect sustained deviations that short recordings or episodic clinic visits systematically miss. White paper
Objective cough trends can function as a remote symptom signal to evaluate real-world treatment response and to detect unexpected symptom trajectories after product launch. This is particularly useful for therapies where cough is clinically relevant. The key requirement is a predefined surveillance question and governance for follow-up and adjudication.
Hyfe is designed for extended monitoring in real-world conditions. Published work using Hyfe includes multi-week monitoring, and Hyfe’s clinical platform is positioned for longer durations when protocols require it - weeks, months or even years.
Hyfe generates time-stamped cough events and derived metrics such as cough frequency over defined intervals (e.g., hourly, daily), monitoring coverage, and trend summaries. The primary unit is an objective cough event linked to a timestamp. This is because Hyfe’s technology monitors cough without recording sound - fully privacy preserving
Cough visualization includes:
Visualizations are designed to support both individual-level review and cohort-level statistical analysis.
Yes. In fact, some leading medical hardware manufacturers are already integrating hyfe’s cough monitoring model - for example Ametris’ category leading LEAP device comes natively with Hyfe’s cough detection.
Yes. Hyfe produces structured, time-stamped outputs that can be transferred into EDC systems, aligned with eCOA timestamps, or integrated into centralized data repositories. Integration pathways depend on sponsor architecture and study design.
Hyfe provides structured exports and API-based access to derived cough data for downstream analysis. Typical formats support standard statistical workflows (e.g., CSV or structured JSON outputs), with documentation aligned to enterprise data requirements.
Yes. Because cough events are time-stamped, they can be synchronized with any other time-series data such as actigraphy, sleep metrics, heart rate, spirometry, environmental exposure, or medication timing.
Dashboards and reports can be configured to reflect study-specific metrics, time windows, and thresholds. Customization typically occurs within predefined data structures to maintain auditability and consistency across participants and sites.
Hyfe supports multiple models depending on use case:
Deployment timelines depend on protocol complexity and integration requirements. For clinical trials, timelines are generally aligned with standard vendor onboarding, data integration, and privacy/security review cycles rather than hardware procurement.
Absolutely. Hyfe’s architecture is compatible with geographically distributed studies, provided regulatory, language, device compatibility, and data transfer requirements are addressed in the study design.
Pricing typically reflects:
Commercial structures are generally aligned with per-participant or per-study models, depending on deployment scope.
Yes. Hyfe collaborates with CROs to facilitate trial integration and with device manufacturers and wearable platforms to enable embedded or hardware-integrated cough detection.
Hyfe integrates at the data layer, exporting structured cough metrics into sponsor EDC systems, data warehouses, or analytics platforms. Integration is designed to align with established clinical data standards and governance frameworks rather than requiring parallel systems.