PeakProfiling: Detecting diseases from the voice
What if you could use the human voice in detecting or predicting diseases? That would allow the health system and patients to be supported in a completely new way. This may seem far away, but PeakProfiling, a sound analytics company, has actually built proprietary technology based on musicology and mathematics to create so-called feature extraction tools that indicate diseases. Its solution is based on decades of research on the characteristics of the voice.
With the expertise of an interdisciplinary team, PeakProfiling conducts hypothesis-driven and exploratory analyses based on classification and machine learning methods to identify disease-specific voice characteristics. The approach is unique since it is rooted in quantitative musicology, combined with AI. The company already built algorithms for detecting diseases such as ADHD, depression, dementia, or ataxia. Additionally, it created algorithms for the detection of softer conditions such as fatigue, well-being, and pain. Together with its clinical, academic, and industrial partners, the company works on patient-centered, objective, and fully digitized solutions.
The benefit is on both the doctor’s and the patient’s end. Firstly, doctors can use it as a supportive screening tool in their clinical practice. Secondly, patients can use telehealth tools, to monitor disease progress without visiting a medical practice. Moreover, clients could use the app to take a 1-minute speech test on the potential risk for different diseases, such as dementia or depression. Consequently, the tool helps health insurers to boost customer acquisition and reduce costs through early intervention.
As a result of the company’s several decades of research, it built a proprietary technology (hundreds of feature extraction tools, partly patented) to extract sound features from the human voice. The approach is rooted in quantitative musicology, which is combined with AI. It developed highly complex features based on hypotheses-driven approaches.
Currently, the tool is mainly working in a B2B context, e.g. with clinics or academic institutions, analyzing already existing data to investigate the relationship between features and different diseases. Additionally, PeakProfiling developed algorithms analyzing softer conditions, such as emotions or fatigue for large corporations.
PeakProfiling has access to sufficient voice data, aiming towards stable algorithms with its proprietary technology to get stable results on real-world data. It is in contact with health insurers to create a business case.
How does it work?
The main focus of the solution is the development of algorithms. Depending on the medical aim, the company translates it to the technical problem and develops hypotheses on how the specific disease might impact the voice and select or reduce features. In the last step, it selects and tunes the best model. As a result, the probability score of having a specific disease is calculated. An applied system could look like the following:
Why did we invite PeakProfiling to present at DIA Munich 2022?
Prevention is an important trend in the insurance industry, no matter the business line. PeakProfiling comes up with an objective and automated analysis of voice to indicate and predict diseases. Combining hypotheses-driven and proprietary voice features with machine-learning methods
At DIA Munich 2022, Dr. Julia Volkening, PeakProfiling’s Medical Affairs Manager, will present the company’s background in quantitative musicology and projects. She will talk about the possibilities of using the technology in health insurance, about some use cases and demo an emotion analysis.
Who is PeakProfiling?
PeakProfiling’s technical founder, Dr. Jörg Langner, conducted research in the field of quantitative musicology in the 1990s, first as a Ph.D. and postdoc at the University of Music and Performing Arts Hannover, and in the early 2000s as a lecturer at Humboldt University, on the quantification of musical phenomena. In 2017 Jörg met Claudio Hasler, CEO and co-founder of PeakProfiling. Claudio worked for Google and Klosterfrau for quite some time and provided strong experience in digital health analyses and business. Together they thought about a digital and fully automated solution using voice sounds to indicate diseases. Thus, PeakProfiling was born and in collaboration with the Charité Berlin, they achieved stunning results in the field of ADHD.
Within the last few years, the team has grown, and PeakProfiling has a very interdisciplinary team from different fields, such as data science, psychology, mathematics, physics, neuroscience and speech pathology. PeakProfiling is fully bootstrapped with revenues.
Within the last years, PeakProfiling won several prizes such as the Health-i Award of the TK and the Handelsblatt, the venture contest of the ETH Zürich, EPF Lausanne, McKinsey and KTI. Moreover, PeakProfiling was selected and successfully participated in the EIT health – startups meet pharma program, Biogen – Neurotechprize, Insurtech Hub Munich – H+ Digital Health, UCB – Digital Care Transformation Incubator and Microsoft for Startups.
“Imagine you detect a disease years before it becomes severe by a simple phone call…or you spot a defect in a machine before a crash happens. The impact we will have with our approach is truly fascinating”Claudio Hasler & Dr. Jörg Langner – Founders PeakProfiling