Developing automated hearing aid technology with a mathematical approach
Bart van Erp defended his PhD thesis at the Department of Electrical Engineering on June 12th.
By 2050, one in four people globally will experience hearing loss due to aging populations. When untreated, hearing loss can lead to significant social isolation, health complications, and economic hardship. Effective hearing solutions become increasingly critical because of this, but for many users the current hearing aids don鈥檛 work adequately for their individual needs in different environments. In his PhD research, Bart van Erp introduces a mathematical approach to hearing aid technology that can take away current limitations and allows users to personally control how they want to experience their acoustic environment.
Traditional hearing aids amplify all sounds uniformly, often failing to distinguish between important speech and background noise. This makes these devices particularly ineffective in noisy places where users need them most, such as restaurants or crowded spaces. Bart van Erp focuses in his PhD research on developing hearing aids that learn and adapt to each person鈥檚 unique hearing preferences and environments. These next-generation devices can automatically recognize whether you're in a quiet office, a noisy restaurant, or at home watching television, and adjusts accordingly. When you adjust settings for specific situations, the technology learns from your interactions and remembers these preferences for similar future environments. This creates a personalized listening experience without requiring technical knowledge from the user.
Probabilistic modeling
To make this technology possible, Bart van Erp uses a mathematical approach called probabilistic modeling that allows hearing aids to separate different sound sources like voices, music, or background noises. The system then balances exploration of new settings with exploitation of what鈥檚 already working well. When encountering completely new acoustic environments, the technology can adapt on the fly, making educated guesses about optimal settings before refining them based on user feedback.
Streamlining mathematical models
A significant challenge while developing these devices was to make sophisticated algorithms work within the severe power and processing constraints of hearing aids. The research demonstrates how complex mathematical models can be efficiently streamlined without sacrificing performance, making them practical for real-world implementation in hearing aids. The next steps involve further refinement and industrial adaptation, with promising opportunities emerging at the intersection of model-based approaches and machine learning that could accelerate the journey from laboratory to life-changing technology.
Fundamental shift
This approach represents a fundamental shift in hearing aid design. Rather than treating all hearing loss the same way, these devices recognize the deeply personal nature of how we experience sound. For users, this is a significant step towards hearing aids that understand and adapt to their individual needs rather than being forced to adapt to technological limitations.
Title of PhD thesis: . Supervisors: Prof. Bert de Vries and Dr. Wouter Kouw.