Real-Time Sound Attenuation for Hearing Sensitivity

A new AI-powered mobile system can selectively filter out specific, bothersome sounds in real-time while leaving the rest of the environment audible. The system, called Sona, was developed by researchers Jeremy Zhengqi Huang, Emani Hicks, and Sidharth to address a key limitation of noise-cancelling headphones for people with noise sensitivity. While active noise cancellation reduces discomfort by suppressing everything, Sona allows users to maintain awareness of their surroundings by targeting only the sounds that cause distress.

From Broad Cancellation to Selective Mediation

For individuals with conditions like hyperacusis or misophonia, the standard tool for managing overwhelming environments has been noise-cancelling headphones or earplugs. These solutions, however, create a significant trade-off: relief comes at the cost of auditory isolation. This can hinder communication, reduce safety awareness, and limit social participation. The researchers aimed to move beyond blanket suppression to create a “soundscape mediator”—a system that intelligently reshapes the acoustic environment in real time.

Sona’s development was directly informed by a formative study involving 68 individuals who self-identified as having noise sensitivity. Their feedback highlighted the need for a system that could handle multiple, overlapping sounds common in real-world settings like restaurants or public transit, and one that could be personalized to an individual’s unique triggers.

How Sona’s Target-Conditioned AI Pipeline Works

The technical core of Sona is a neural network pipeline designed to run efficiently on a mobile device. Its primary advancement is overcoming the “single-target” limitation of prior audio filtering research. Earlier systems could typically isolate or remove only one type of sound at a time. In contrast, Sona’s “target-conditioned” architecture allows it to process and simultaneously attenuate several different sound sources from a mixed audio stream.

The process works in real time: the system continuously analyzes incoming microphone audio, identifies which sounds belong to user-selected classes (e.g., chewing, traffic noise, construction), and generates an output audio signal where those specific sounds are reduced. All other sounds, including speech and ambient cues, are preserved. A key feature is user extensibility. If a person is bothered by a sound not in the default library—like a specific appliance hum—they can teach Sona by providing a few in-situ audio examples. The system adapts without requiring a full, computationally heavy retraining process.

Technical Performance and Real-World Testing

The team conducted rigorous technical benchmarking, confirming that Sona operates with low enough latency (delay) to be suitable for live listening, a non-negotiable requirement for a wearable aid. More importantly, they evaluated its effectiveness through an in-situ study with 10 participants who used Sona in their daily lives.

Findings showed that the system enabled meaningful reductions in the intensity of bothersome sounds as perceived by users. Critically, participants reported maintaining a significantly greater awareness of their surroundings compared to using conventional noise-cancelling headphones. This balance is central to the system’s value proposition.

Practical Implications for Hearing and Sound Sensitivity

This research represents a shift from passive sound blocking to active, intelligent sound management. For clinical populations, tools like Sona could serve as a complementary aid. It does not treat the underlying neural mechanisms of conditions like tinnitus and hyperacusis or misophonia, but it offers a novel way to manage environmental triggers. By reducing the acute stress of sound exposure, it may help individuals engage in activities they would otherwise avoid, potentially supporting broader therapeutic goals like desensitization or improved quality of life.

The user-extensible design is particularly significant for misophonia, where triggers are highly individualized and often relate to human-generated sounds. The ability to personally train the system aligns with the need for personalized management strategies, as discussed in resources on misophonia coping. It empowers the user to define what constitutes a “bothersome” sound in their own environment.

Toward a New Class of Personal Acoustic Assistants

The work by Huang, Hicks, and Sidharth points toward a future where personal AI systems act as mediators between individuals and their sensory environments. The goal is not to remove a person from their environment but to reshape that environment to make it more livable. This supports both comfort and the fundamental human need for social connection and situational awareness.

Future iterations could integrate with hearing aids or augmented reality glasses, creating more seamless experiences. Further research will need to explore long-term effects, usability for diverse populations, and integration with other therapeutic approaches. For now, Sona provides a compelling proof-of-concept: that with on-device AI, we can begin to craft our personal soundscapes in real time, moving closer to a world where noise sensitivity does not necessitate social isolation.

Source: Research by Jeremy Zhengqi Huang, Emani Hicks, and Sidharth. For full details, refer to the original paper via its DOI or PMID identifier.