Frequently Asked Questions
Hyfe is an artificial intelligence company devoted to pioneering the field of acoustic epidemiology, and its first focus is on cough. To support researchers in this field, Hyfe has developed a set of cough monitoring tools (the "platform"). Currently these tools include a consumer-facing smartphone app (simply named “Hyfe Cough Tracker”) for both iPhone and Android, a research-facing Android-only app tailored for scientific studies (“Hyfe Research”), an app for people to “label” sounds (“Hyfe Labeling”), an interactive dashboard for managing an ongoing study (“Hyfe Research Dashboard”), and a web-based screening app for high-throughput testing sites (“Hyfe Screening”).
Researchers use Hyfe’s apps in a wide variety of ways. The most common research applications we have seen are:
Cough as a biomarker: When participants come to clinics for medical tests, their coughs are collected and paired with their medical tests. When participants return for subsequent tests (e.g., after the onset of treatment), cough is collected again to produce a longitudinal cough dataset.
Remote cough monitoring: Between initial and follow-up visits to a clinic, participants are given a study phone with the Hyfe Research app installed. This phone captures all coughs during the monitoring period. They are asked to carry this phone with them until they return.
Inpatient monitoring: Phones are placed next to hospitalized patients to track their coughs over the course of their stay.
Cough screening: Cough collection is integrated into a medical screening center (e.g., COVID-19 testing site) using either the Hyfe Research app or the Hyfe Screening website, allowing researchers to pair participants’ coughs with other test results.
Syndromic surveillance in a population: Participants download Hyfe Cough Tracker on their own phones and go about their lives. De-identified locations are tracked by the app, if users opt in.
Veterinary medicine and livestock monitoring: The Hyfe Veterinary application (not yet released) is placed in an area with one or more animals. Coughs are quantified over space and time.
The studies enabled by these forms of data collection also seem to be limitless. For example of what can be learned from these research techniques, visit our Publications page (link).
Once the app is installed, the user opens the app, signs in, and carries out an ‘onboarding’ routine to ensure that the phone can recognize their cough. The app operates in the background to listen continuously for explosive ‘peak sounds’ that might be a human cough. When a peak sound is detected, the app saves a 0.5 second snippet of the sound. That snippet is passed to Hyfe’s machine learning algorithms to classify it as a cough or not.
Here is an example of how the Hyfe Screening website works. In short, the website displays instructions for the participant to cough at the end of a 5-second countdown. The website then listens for a cough, then confirms that a cough was successfully collected.
In studies using the web-based screening app, Hyfe will give each participant a unique URL to the screening app; this URL contains the participant’s Hyfe user ID, so that the sound data collected can be linked to that specific participant.
The Hyfe apps are designed to perform well across all devices and in most soundscapes, and they are continually being improved. Hyfe is committed to providing updated performance statistics on a quarterly basis beginning in July 2021, and it supports research groups in conducting field trials that allow them to understand the performance expected for each specific study.
It is important to note that performance can vary across phone makes and models, which have microphones of differing quality and different standards for sound processing, and it is compromised by very noisy conditions. For more information, see the FAQs under the “Operational Logistics” section below.
Hyfe takes user privacy extremely seriously. The app listens only for explosive sounds and saves only 0.5-sec snippets for analysis. The only exception to this policy is that, for consenting participants using the Hyfe Research app, de-identified 30 sec files are randomly saved to use for internal data quality analysis (these files are used to ensure that our apps are not missing any coughs). User locations in the "Hyfe Cough Tracker" app are collected if the user opts in, but these locations are de-identified by restricting accuracy to 0.5 km. Hyfe does not share any data with outside entities other than research partners, and research data are already de-identified.
The Hyfe Research app is designed to ensure confidentiality; once installed, the user signs in with a pre-determined user ID (e.g., UG623) and password that is provided by the researcher. This allows the researcher to couple cough data to the participant’s other data in a way that protects the participant’s identity in the Hyfe database.
Anyone can download and use Hyfe. For academic or medical research, ethical approval is required (since any research on human subjects typically requires IRB approval). Hyfe has a library of IRB materials from which researchers can draw. Examples of this content are available on the IRB and Grants Info page.
Hyfe is committed to making its research tools accessible to everyone who wishes to use them to generate meaningful knowledge. Once you submit a request to begin a research cohort, Hyfe will contact you to discuss the cost.
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 will generate a research cohort for you with a discrete number of user IDs - password combinations, and it will create an administrative account that allows you to manage your project using the Hyfe Research Dashboard.
This will depend on the details of your study, but general instructions for installing the Hyfe Research app and enrolling patients can be found in the Hyfe Research Users Guide.
You can manage and monitor your project using your account on the Hyfe Research Dashboard. From the dashboard you can intermittently request data reports to be prepared for you by the Hyfe team.
You will have access to all acoustic recordings and user data for your project. You may also have access to disease prediction scores (e.g., TB, COVID-19, etc.), depending on the details of your arrangement with Hyfe.
Hyfe considers its partnerships with researchers to be symbiotic: Hyfe enables partners to conduct impactful research, and its partners help Hyfe collect the coughs that train its detection and classification models. Typically, partners and Hyfe will share joint ownership of the data. The partners have unrestricted use of the acoustic and user data from your project for whatever purposes you wish. Likewise, Hyfe will have unrestricted use of the acoustic and user data for building and improving its technology.
Hyfe manages and stores all of the cough- and app-related data and provides cough prediction scores for each explosive sound detected by the app. All these forms of data management and analysis are services that are common to all Hyfe research partnerships. Additional services such as disease prediction scores, human labeling of sounds, data visualization, and more advanced analyses may also be included in your research agreement with Hyfe.
Absolutely! Hyfe has developed an open-source R package, ‘hyfer’, devoted to empowering researchers to analyze their data on their own. This package is set for public release in summer 2021.
Not necessarily; Hyfe offers this service for a small cost.
This is definitely a possibility, and we can work with you to determine if and how you can implement your own labeling workflow for your project.
Most studies will be best served by the Hyfe Research app, but it depends on your objectives.
Here are some considerations:
This depends on the setting, your study population, and your objectives. Here are some considerations:
Hyfe Research is designed to work on any Android smartphone, and the consumer-facing Hyfe Cough Tracker is designed to work on any Android or iPhone. That said, certain phone manufacturers collect higher-quality sound data than others. The only problematic manufacturer we have encountered is Nokia; for the time being, these should be avoided if possible.
In the consumer-facing Hyfe Cough Tracker app, there is a levers button in the upper left of the main page. If you click it, you’ll see recording status. For the Hyfe Research app, you can visit your Research Dashboard and check the status of each phone.
Hyfe works with headphones. On most devices we have tested, the app will not record while using its phone functionality.
The Hyfe Research app does record 24/7. The only way to pause recording is to close the app. The consumer-facing Hyfe Cough Tracker app allows you to pause listening or set up your phone for nighttime-only listening.
We have found that app sensitivity declines significantly when users are 2 or more meters away (7 feet). In one sense this is an advantage, since it reduces the number of coughs accidentally detected from those other than the user. This can also be a disadvantage if participants are not used to keeping a phone close to them at all times. We encourage users to keep phones within arms reach when they are at home.
Hyfe invites all researchers to carry out their own tests with their own study phones to gauge the range and sensitivity of the app. Instructions for these field tests are provided here.
Currently no, but this is a feature that we are developing now.
The Hyfe apps are designed to work without special carrying behaviors or auxiliary microphones. We believe this is the strongest aspect of our product: we capture coughs in the real-world. However, participants who are not used to carrying a phone can benefit from certain carrying devices such as armbands (e.g., link), necklace holders (e.g., link), or waist bags (e.g., link). To date we have found that armbands and necklace holders perform best (with the phone upside down, so that the microphone faces up towards the user’s mouth).
Yes, though performance is compromised slightly. We have found that it helps if users remember to keep their phones upside down when carried in a pocket or purse.
There are three steps you can take to assess the quality of your cough data. First, track your data collection on the Hyfe Research Dashboard. Second, carry out brief tests to get a basic sense of how the app will perform. Instructions for these field tests are provided here. Third, actively label the peak sounds detected from your study phones to gauge how well the app is performing. Alternatively, you can request a Performance Report from Hyfe (this option will soon come online in the dashboard) that will use labelled sounds as well as labelled sessions to determine rates of misclassification.
The app does reduce battery life slightly, but this impact is being continually improved with subsequent versions of the app.
Yes. The acoustic and user data are stored internally until the phone connects to WiFi, at which point it is uploaded to the Hyfe cloud servers.
This greatly depends upon the user’s cough patterns and the noisiness of the user’s setting, since every explosive sound is saved and analyzed. On average the app tends to use 1.0 - 1.5 GB every 10 days.
When the phone’s internal hard drive reaches less than 5% available storage, the Hyfe app will stop recording. When this happens, a notification will appear saying that this has occurred. Once more than 5% storage becomes available, the app will begin operating again.
No. At this time the apps only support internal data storage.
If the phone has an active SIM card and has granted the Hyfe app to use cell data, the app will send data to Hyfe’s servers whenever it has signal. Otherwise, it will send data to Hyfe’s cloud servers whenever it is connected to WiFi. If the phone has been without service or WiFi for several days, it could take minutes to hours to upload all of the app data.
Yes. As with any app, you can control which apps are allowed to use cell data using your phone’s settings.
Sounds will continue to be stored locally until an internet connection is re-established.