Frequently Asked Questions
Hyfe is an artificial intelligence company pioneering the field of acoustic epidemiology. Hyfe is the global leader in cough technology and cough science. Hyfe's cough detection technology performes as good or better than a trained human ear in the real world (get in touch and ask for validation paper) and our data insights platform decodes health signal from cough frequency. These longitudinal cough monitoring and insights capabilities are used daily by thousands of researchers, pharma, physicians and patients across the world. From remote-patient-monitoring in chronic care management, to digital biomarkers for clinical trials, Hyfe brings precision medicine and personalized approaches to respiratory health.
Hyfe’s AI models are trained on the world's largest dataset of cough sounds, and have processed billions of datapoints across 160+ countries. Hyfe is also being used in 46+ research trials across indications including asthma, COPD, Congestive Heart Failure and Covid and has been featured in more than a dozen publications.
Tell us about your project by submitting a request to begin a research cohort. Hyfe will then contact you to discuss arrangements and share instructions for getting started.
Hyfe has redefined the way that cough is quantified, offering precision, convenience, and scalability. By using powerful AI models that run in the background and preserve privacy, Hyfe’s tools can detect cough passively, as it happens. These tools are designed for the real world and require minimal infrastructure, and do not require active participation from users to function. Hyfe’s cough monitoring capabilities can be deployed on any smart device with a microphone, allowing cough monitoring to be used at any scale.
By leveraging acoustic AI that can run on any smart device, Hyfe enables patients to track their coughs, providers to use cough as an objective clinical finding, and researchers to use cough as an endpoint for clinical trials. This represents a major breakthrough in the fields of precision medicine and in respiratory health and wellness.
The most common alternative approach to AI powered cough analysis is the single-cough analysis model, in which a patient is asked to provide a sample of a single or a few elicited coughs. These samples are then analyzed for structural signals (in some literature called “acoustic fingerprints of a cough”). There are significant limitations to this approach, including biological plausibility, infrastructure, costs and scalability issues. Here is a detailed analysis of the cons and pros of both AI cough analysis frameworks.
Hyfe's models are trained on tens of billions of data points, collected from actual coughers, across diverse environments and demographics, ensuring high performance and real-world applicability. All of Hyfe’s training data is collected in the real world, across +160 countries and across every possible socio-demographic segment. It is also collected on every possible hardware / software / acoustic environment combination, making it the world’s most extensive cough sound dataset, and ideal for training high performance acoustic machine learning models.
Additionally, Hyfe has been written into more than 46 IRB approved research trials, allowing the processing of millions of medically validated data points as well as additional medical ground truth data including socio-demographics, disease status, radiology reports and more. This allows Hyfe’s AI models to be trained on extraordinarily high quality data, leading to a category leading performance in the real world.
Continuous cough monitoring models - such as the ones developed by Hyfe - are fully privacy preserving, as they run on device and no sound data gets recorded or processed outside of the device on which the model runs. The key advantage of such on-device AI processing is the enhanced privacy it offers. This is particularly important in an era where data breaches are common, and sensitive information is highly valuable. Users can interact with these models with the confidence that their data, behaviors, and preferences remain confidential and are not exploited for advertisement or any other purposes without their consent.
Additionally, on-device processing inherently boosts security. By minimizing data transmission over networks, the risk of interception by unauthorized parties is drastically reduced.
This is in stark contrast with AI single-cough classifiers which, by their nature, require a deep analysis of the cough sound (its acoustic “fingerprint”) itself. This process has high computational requirements and is done in a cloud set-up. When sound recordings leave the user's device, it significantly increases the risk of personal information being intercepted, leaked, or misused by third parties. Also, given the nature of the analysis, a cough thus processed might be a plausible biometric marker that might be later used to uniquely identify a person.
Because of powerful, yet relatively simple cough / non-cough classifiers, Hyfe’s models run on-device, using edge computing. This means that no identifiable data ever leaves the device, which enhances privacy and security, and eliminates data transmission and manipulation risks that are common with other acoustic or imaging AI models.
Continuous cough monitoring can be deployed at scale via any smart device, and has no infrastructure dependencies. This makes it especially suitable for low infrastructure settings. It offers a powerful tool for public health surveillance and disease outbreak detection as well as a proven method to screen for diseases including TB, Malaria and others, in real time, at scale.
Hyfe’s cough detection and monitoring technology is device agnostic, working with any device that has a microphone, even a basic one. This includes every smartphone on the market, most wearables and other smart devices. The technology runs on edge, fully on device, so not even an internet connection is required for the basic detection models to run. Hyfe’s insights API requires an internet connection - even a very slow one will do.
For the purposes of research and life sciences, Hyfe has a proprietary, fully optimized wearable device that we make available to our partners. Get in touch to learn more.
Hyfe’s cough-monitoring technology has been used across all age groups. The dataset the AI models are trained on is very large and diverse, which ensures cough-monitoring is accurate across al types of demographics. However please note that the FDA application we have made is for cough-counting in adults only.
Hyfe’s models are trained on an extensive and highly heterogenous (diverse) dataset. Over the years Hyfe’s models have processed hundreds of billions of datapoints and many centuries of continuous person-time monitoring.
The models are trained on real-world data across +160 countries and they also benefit from millions of datapoints of medically labeled data. Hyfe has completed rigorous validation trials that demonstrate a newr-perfect performance in the real world - please get in touch and request the validation data if interested
Advanced algorithms and a substantial training dataset help minimize errors. Hyfe’s models have an extremely high performance in the real world. With negligible rates of false positives or negatives. Additionally, the signal in cough monitoring comes from cough patterns over time and because of that, even an occasional false positive or negative would normalize over time - the way a step counter normalizes false positives and negatives.
Unlike single-cough classifiers, which require the control of variables such as hardware, software and acoustic environments, Hyfe’s cough detection and monitoring models are trained on real world data and designed to run on any device and in any reasonable acoustic environment. Because of this there are very little limitations to scaling this technology fastly and cheaply.
Data privacy and security are critical concerns when using AI tools. Mindful of the importance of privacy, Hyfe's on-device processing models ensure enhanced privacy and security, with minimal risk of data breaches or unauthorized access.
Continuous cough monitoring allows for the establishment of personalized baselines and dynamically adjusts to individual coughing patterns, offering customized health insights. This is in stark contrast with single-cough analysis models that cannot establish individual baselines may fail to recognize important changes in a patient’s cough.
Future developments include increased sensitivity through additional data points, integration with smart home devices, and broader applications in personalized medicine and public health.
One very significant development - that Hyfe is in the process of bringing to market already - is behavioral cough suppression therapy that uses individual cough patterns to inform a personal therapy regimen for chronic coughers. This approach is based on rigorous scientific evidence that shows an efficacy of almost 80% for behavioral cough suppresison therapy for patients with cough hypersensitivity/ chronic cough. LINK
Hyfe's advanced AI models are designed to minimize the impact of external noise. While Hyfe has developed experimental models that distinguish between different people’s cough, these models have to be used with caustion for at least one critical reason: due to the biological nature of cough, an individual’s cough can change because of a specific disease. Building individual cough detection models at scale could mean that the model would fail to detect coughs exactly when they are needed.
However, analysis of extensive real-world cough datastreams done by Hyfe demonstrates that the presence of others' coughs does not significantly damage the overall data quality or insights derived from it.
Acceptance of Hyfe's data by the FDA would depend on the context of its use and the quality of the validation studies submitted. For clinical applications or claims, rigorous validation and regulatory review is necessary and Hyfe has completed a rigorous round of validation studies (we can share on request). Hyfe's approach to continuous cough monitoring, coupled with its potential for healthcare impact, positions it well for pursuing regulatory approvals
Hyfe is certified as a CE Mark 1 device in Europe and has submitted an application for FDA certification in the USA. It is expected that Hyfe’s Cough Monitor will be the first-ever regulated cough monitoring tool.
Meanwhile, regulatory clearance is not required for the use of cough detecting technology as part of wellness products and it is not required for the use of this technology in research settings.
Evidence suggests that tracking cough can predict exacerbations in conditions like COPD, asthma. CHF and even detect early signs of diseases like lung cancer. Hyfe's technology, with its capability to monitor cough dynamics continuously, supports the notion that cough monitoring is a valuable tool in disease management and early detection.
Even before obtaining FDA clearance, Hyfe can offer its technology for general wellness and monitoring purposes, providing insights into cough patterns and frequency changes without making specific health claims that require regulatory approval. This means Hyfe’s tools can help monitor coughs and say when a person’s cough-rate is statistically changing, but no further commentary or call-to-action.
Hyfe's continuous cough monitoring technology is designed for ease of integration into any existing platform. Given its technical simplicity and the potential for deployment via small SDKs, integration is straightforward and should not take more than a few days. However, it all depends on the specific systems and requirements of the platform.
Get in touch for a concrete estimate
The cough detection models create timestamps for every cough detected. These timestamps are fed into Hyfe's proprietary insights APIs, which analyze them and build the insights dashboard.
The only data processed is text based - timestamps.
Hyfe's detection models run on device, using a technology called edge computing. This means that detection and processing happens on device and no data leaves the device.
The way the models detect cough is they listen for a specific type of sound called a "peak". When a peak is detected, this peak gets classified as cough or no cough. If the peak is a cough, a datastamp is issued, which means the exact time and date of the cough is recorded. The timestamp is the only piece of information recorded.
There is a whole new generation of single-data screening methods using acoustic AI. In the typical such model, a patient coughs once into a device in order to receive an accurate diagnostic. While exciting technically, this approach faces significant limitations - as detailed here. Some of these limitations are intrinsic to the biological nature of cough. Others are related to the training dataset quality, and variability introduced by the need to control hardware calibration, software, firmware, and acoustic environments when collecting acoustic samples. A similar need to control noise is also required during the model’s training, creating an entirely new problem - and risk category - for scientists building single-cough classification models, which are not generalizable in the real world environments.
While there isn't any literature confirming the viability in practice of single cough analysis models, there are several publications showing lack of accuracy and low performance in the real world - here is a recent example
Finally, regulatory frameworks would make it very hard for such single cough screening technology to scale.
Continuous cough monitoring emerges as a vastly superior alternative, offering a scalable, cost-effective, and non-invasive solution that has very limited infrastructure dependencies and a very low variability related to hardware/ software combinations or acoustic environments. Continuous cough monitoring can significantly enhance patient outcomes and reduce healthcare costs by detecting new onset respiratory infections, predicting exacerbations, indicating objective treatment or disease progression dynamics, and resulting in reduced hospital readmissions. There is already mounting scientific evidence that cough frequency correlates with disease status in TB (ref and ref) as well as Covid19.
Considerations include ensuring high accuracy in diverse environments, minimizing hardware / software dependencies, and addressing clinical validation challenges to meet regulatory standards.
These considerations are significant factors in continuous cough monitoring being a superior approach to cough evaluation, versus single-cough analysis models.
Hyfe has +90% sensitivity (i.e. detects 90% of coughs) with just 1 false positive per hour, in real-world environments while patients go about their normal daily life.
The early detection and management of respiratory conditions have long been a challenge in medical sciences. Traditional screening methods have always been high friction, cumbersome and expensive, in addition to having low sensitivity.
Continuous cough monitoring emerges as a superior alternative, offering a scalable, cost-effective, and non-invasive solution that has very limited infrastructure dependencies and a very low variability related to hardware / software combinations or acoustic environments. Because it runs passively and continuously, it is an ideal approach to screening as it can detect subtle changes in cough patterns before further symptoms develop, and long before the patient becomes self-aware. Continuous cough monitoring can significantly enhance patient outcomes and reduce healthcare costs. By detecting new onset respiratory infections, predicting exacerbations (e.g., COPD) and decompensation (e.g., Congestive Heart Failure), indicating objective treatment or disease progression dynamics, adverse drug reactions (e.g. cough due to ACE inhibitors or cough in drug-induced interstitial lung disease), it results in timely interventions, reduced hospital readmissions and better prognosis for the patients.
Hyfe’s cough detection models output a series of timestamps - a small, simple datastream - which a secondary Insights AI consumes and analyses continuously. The detection model is device agnostic - it leverages any standard microphone - in a smartphone, wearable, hearable, headset, bedside computer or any other modern smart device - to passively collect cough data over extended periods.
More on the differences between continous cough monitoring and single-cough analysis
Hyfe AI models are optimized for continuous cough detection and analysis of cough patterns. Direct disease diagnosis would require further indication-specific clinical validation and regulatory approval. The existing cough monitoring capabilities are extremely valuable as a tool aiding in diagnostics as it can indicate changes in cough frequency and patterns that are consistent with certain indications. There is peer reviewed evidence indicating the value of Hyfe’s technology in screening for TB or predicting outcomes in COVID and there is mounting evidence that cough monitoring can be an effective early warning system for a host of other diseases, including ILD, COPD, Asthma, Lung Cancer and many more.
Hyfe’s models look for signals through quantitatively analyzing cough frequency dynamics and patterns rather than the acoustic qualities of any individual cough (acoustic “fingerprints”). This minimizes the impact of external variables and eliminates biological “unknowns” related to the nature of cough, focusing on the emergence of patterns that indicate the onset or exacerbation of a health condition. It also allows for individual customization and personalized baseline levels - for patient A it could be “normal” to cough 65 times a day, while for Patient B, 18 coughs is above baseline.
Even once a baseline cough rate for Patients A and B are established, being a stochastic symptom, cough is highly variable for each of them hour-to-hour, day-to-day. Coughing is, however, not random and cough counts follow statistical distributions. This means that the only reliable way to ascertain whether Patient A’s cough is signaling a change in health / disease state is to measure it continuously, over a sustained period of time and operate with dynamic thresholds that are custom to each user and moment in time.
The technology has already shown impact, with evidence suggesting that COPD exacerbations can be predicted days in advance to hospitalizations by monitoring subtle changes in cough frequency via a stationary bedside device [ref]. Similarly, the persistent, low-key cough associated with lung cancer, often ignored by patients, can be detected early, increasing the likelihood of successful intervention. Read more about cough monitoring in lung cancer here.
While several teams are working on single-cough diagnostics tools - in areas that include Covid19, Tuberculosis, Sleep Apnea and other indications - the relevant literature does not show sufficient evidence that this is possible.
In addition to the scientific doubts, there are significant technological and operational barriers to scaling any single-cough analysis model - see here for more details.
Yes, cough monitoring complements many other other diagnostic methods or diagnostic support tools, offering a holistic approach to patient health monitoring and management. Indication-specific sensitivity increases significantly when two or more datastreams are analyzed together - for example cough rate and fever for infectious disease.
Continuous cough monitoring is a technology that passively and continuously detects cough and its dynamics, providing insights into a patient's respiratory health. Continuous cough monitoring has been pioneered by Hyfe and it represents a groundbreaking advancement in healthtech and a significant step towards bringing precision medicine to respiratory health and respiratory wellness. The ability to monitor cough continuously is changing our approach to screening, diagnosis, prognosis, and management of a wide range of acute and chronic clinical conditions across respiratory, cardiovascular, infectious and other fields, including but not limited to:
Unlike single-cough analysis, which analyzes individual cough samples for disease indicators, continuous cough monitoring tracks cough frequency and patterns over time. This longitudinal approach allows for the detection of changes relative to an individual's baseline cough frequency, offering a more dynamic and comprehensive understanding of their respiratory health.
Single cough analysis - as well as cough analysis conducted on several elicited coughs is an approach, where a patient provides one or several cough samples and a machine learning (ML) model analyzes these samples to look for weak signals that would be consistent with a certain trait or feature, e.g., a disease.
Given the biological nature of cough, this requires that the samples are collected with as little variability as possible from the data that was used to train the model. This means at least the same hardware, software and a clean acoustic environment. Other variables that may affect the performance of this approach include uncertainty around disease-consistent structure of a cough, across age, gender, social-demographic groups and diseases as well as the need for a patient to initiate the screening.
Continuous cough monitoring tools - such as those developed by Hyfe - have a completely different approach that eliminates most of the limitations and risks characteristic of the single-cough approach. They are built around a powerful cough / non-cough AI classifier that runs continuously and thus does not require any action from the patient / user. Cough / non-cough classifiers are a more practical problem to solve than a disease specific classifier and, because they do not rely on potentially weak indication-specific signals within individual coughs, they do not require a controlled acoustic environment or specially calibrated hardware or software. Cough / non-cough classifiers benefit from the scalability and versatility of a software-only model and, as has been proven in research, have very high accuracy in very diverse real-world environments, even when run on basic smartphones or wearables.
AI-powered continuous cough monitoring, such as achieved by the tools developed by Hyfe, uses advanced machine learning (ML) algorithms to analyze longitudinal data streams of cough sounds, identifying cough patterns and cough dynamics in real time. This allows for the detection of subtle variations in cough frequency that could indicate health changes or conditions.
Cough detection in the real world - historically a hard problem in respiratory health - has been solved with advanced AI models, such as those developed by Hyfe, performing in real world environments with a sensitivity comparable or better than a trained human ear <feel free to get in touch with our team and request the validation manuscript if interested in objective performance>.
Continuous cough monitoring is cough detection deployed passively and continuously. AI-powered cough monitoring allows the deployment of cough monitoring capabilities at any scale, fastly and cheaply. Any basic microphone can be utilized for data collection, regardless of acoustic environment or software / hardware combinations in any individual device. This significantly simplifies the technology required and allows for broad implementation across diverse settings, almost independent on any available infrastructure.
By leveraging powerful machine learning algorithms, continuous cough monitoring defines personalized baseline cough levels for individuals and detects significant changes in cough trends in real time. This AI-driven approach adapts to individual variations in coughing, offering personalized health and wellness insights.
Continuous cough monitoring has applications across screening, diagnosis, prognosis, and management of respiratory, cardiovascular, and infectious diseases. It offers a non-invasive, scalable, and cost-effective solution for early detection of conditions, monitoring disease progression, and assessing treatment effectiveness. Continuous cough monitoring models - such as those built by Hyfe - can run on any microphone-enabled device and perform well in the real world. They can be rolled out at any scale fastly and cheaply.
Continuous cough monitoring allows regular people to better understand their health and wellness, and have more insightful conversations with their doctors. It allows healthcare providers to work with objective data from their patients rather than only evaluate subjective recall about cough and its dynamics. Cough monitoring allows for the early detection and prevention of events like COPD or Asthma exacerbations or decompensation in Congestive Heart Failure, and it allows for the more personalized chronic care management.
Finally, cough detection and monitoring has critical application in drug development and biopharmaceutical innovation. By allowing an objective continuous measurement of cough in the context of clinical trials, or academic research models like those built by Hyfe are ushering a new era of innovation in digital respiratory biomarkers and respiratory health.
The easiest way for patients to access continuous cough monitoring technology is by downloading one of the consumer wellness apps that are powered by Hyfe’s powerful AI models on their own smartphones. These include CoughPro and CoughTracker. CoughPro is a full featured premium flagship app with the latest features and exclusive access to a global community of coughers and experts. In addition, CoughPro will soon feature content on behavioral cough suppression techniques addressing cough hypersensitivity. CoughTracker is a more basic, free cough tracking wellness app for both iOS and Android smartphones.
Patients will soon be able to access medical-grade cough monitoring via a dedicated OTC wearable medical device, designed for high precision monitoring and approved by regulators.
Absolutely. While Hyfe's cough detection and monitoring technology runs and performs well on any smart device, in clinical trials settings we recommend the use of a wearable device to maximize data quality over time. Hyfe has a wearable device that has been optimized for clinical research settings as well as a clinical research platform complete with cohort management and dashboards.
Learn more about our work in Life Sciences
Absolutely. Hyfe is working extensively with clinical trials and drug development. A number of biopharmaceutical companies are adding continuous cough monitoring to their development pipelines in stages of pre-screening, screening, as an exploratory endpoint and for post-market activities.
Absolutely. While Hyfe's cough detection and monitoring technology runs and performs well on any smart device, in clinical trials settings we recommend the use of a wearable device to maximize data quality over time. Hyfe has a wearable device that has been optimized for clinical research settings as well as a clinical research platform complete with cohort management and dashboards.
Learn more about our work in Life Sciences
Absolutely. Hyfe is working extensively with clinical trials and drug development. A number of biopharmaceutical companies are adding continuous cough monitoring to their development pipelines in stages of pre-screening, screening, as an exploratory endpoint and for post-market activities.
You can request access to the Research Package and use our online pricing calculator to estimate the cost for using CoughMonitor in a clinical trial. Please note the cost are only for academic research, and include the following:
Analysis services are not included in this pricing. These costs vary depending on the type of analysis needed for the study. Hyfe will provide you with a customized quote.
Yes! Hyfe has a dedicated team that provides excellent analysis services
CoughMonitor Suite is Hyfe’s research product and consists of:
Hyfe’s cough detection and monitoring tools have been used extensively in research - more than 46 research trials across more than 12 clinical indications, resulting in 13 peer-reviewed publications to date. Hyfe has a dedicated program for researchers, specifically designed and priced for investigator-initiated research.
Click here to learn more and download the research package.
Researchers have access to cough frequency data as well as user data for your project. Hyfe provides timestamps of each detected cough sound, and generates longitudinal continuous cough frequency data, accessible in near real time. You may also have access to additional information, depending on the details of your arrangement with Hyfe. Hyfe's research platform offers extensive control to investigators and can also be deployed in blind as well as double-blind randomized control trials. Researchers can use smart devices as well as fully optimized wearable devices made available by Hyfe.
Learn more about our work with researchers
Hyfe has a library of IRB materials from which researchers can draw. Request access to the Research Package to see examples of this content.
Hyfe's cough detection and monitoring technology is optimized for field research at any scale. It has already been used and validated in more than 46 investigator-initiated trials, across 5 continents. It is also used in clinical settings as well as in life sciences settings.
If you are intersted to use Hyfe for research purposes please click here
Researchers use Hyfe’s technology in a wide variety of ways. The most common research applications we have seen are: