Startups use deep learning and advanced signal processing to improve hearing quality.
Machine learning (ML) has quietly become part of our daily lives, powering our cars, the operations in our hospitals and the food we eat. Some of the most exciting companies applying ML don’t even call themselves “ML” companies. Auto is the most obvious example, as $10 billion of funding poured into the industry in 2019 alone. As we look forward to the next generation of verticals where ML will have a revolutionary impact, healthcare stands out as one the most promising. Within healthcare, there are now remarkable ways that ML can improve the quality of our hearing.
A Centuries-Old Problem
The quietest sound you can hear starts to deteriorate as you age, but the loudest audible sound remains constant throughout your entire life. Hearing aids traditionally work by amplifying sounds with only very coarse adjustments, from the clinking plates in a restaurant to the sound of your child's voice. They can be incredibly painful to wear, often drowning out the sounds you want to hear with noises you don’t and this is particularly salient in noisy public places.
There are an estimated 48 million Americans experiencing some degree of hearing loss. Despite the prevalence of the problem, the predominant hearing aids on the market today were built 50-100 years ago. Sonova is a US$17 billion Switzerland-based company that was founded in 1947, the same year the transistor was invented. Demant is a US$8 billion Denmark-based company that was founded in 1904, predating even the Model T. A handful of new entrants recently emerged, offering less expensive, direct-to-consumer products, but achieving quality at par with industry incumbents has remained elusive.
Enter the Age of Machine Learning
With ML, companies can apply cutting-edge technology to transform an age-old problem. Startups are leveraging deep learning and advanced signal processing at a granularity not previously possible to improve hearing quality.
Some incumbent hearing aid companies have recently touted their ability to add “AI” features such as Alexa integrations and step counters. Unfortunately, these features don’t seem to improve actual hearing quality nor take advantage of true ML capabilities beyond generating marketing buzz.
In contrast, startups are building software-based solutions that can continuously ship new features and upgrades to existing users, much like the systems pioneered at Apple and Tesla, resulting in a product that keeps getting better over time. The algorithms that power these hearing aids can detect, predict and suppress unwanted background noise, forming the basis of a non-trivial technical moat behind the product. Engineers have now spent years building these neural network models by taking structured and unstructured data, augmenting it with in-house data (representing a spectrum of ages, languages and voice types), feeding the data into neural network training and then refining the predictions - a process that continuously improves the efficacy of the product.
In my conversation with Andre Esteva, the Head of Medical AI at Salesforce, he noted that “traditional approaches have been limited by extensive manual efforts to acquire data, hand-craft it into a usable format, prepare rudimentary algorithms and deploy them to devices. In contrast, ML has a natural flywheel effect in which devices collect data at scale, ML training protocols automatically process the data, update themselves and redeploy. The effect is a significant reduction in product feedback cycles and an increase in the range of capabilities available. The beauty of this approach is that the underlying intelligence improves over time as the neural nets go through iterative training.”
Time to Invest is Now
There are many reasons to be optimistic about the value ML can create to improve our health in the years to come. I strongly believe that among the next generation of industries where ML will also have a revolutionary impact, healthcare is one the most promising. The future of hearing, primary care, clinical trials and many other critical pieces of the healthcare puzzle will increasingly rely on intelligent systems. Exciting startups to emerge in the hearing aid market are just one example of a new wave of ML-driven innovation in healthcare.
