The widespread adoption of wearable audio devices in recent years has been significantly driven by advancements in active noise control (ANC) technology. ANC allows users to enjoy high-quality audio content even in noisy environments by effectively suppressing unwanted ambient sounds. However, while this technology enhances the listening experience, it can also significantly reduce situational awareness, making it difficult for users to perceive critical environmental sounds. This diminished awareness may delay their ability to react to urgent auditory cues, such as approaching vehicles, alarms, or emergency sirens.
Concurrently, Sound Event Localization and Detection (SELD) systems have been developed to detect, classify, and localize sound events, providing users with crucial awareness of their surroundings. These systems are capable of identifying critical soundssuch as car horns, sirens, or alarmsand accurately determining their direction in real time. By incorporating Head-Related Transfer Functions (HRTF) into the pipeline, SELD systems can spatially encode detected sound events, effectively replicating their natural directional perception. This spatial representation ensures that users remain alert to potential dangers while continuing to enjoy their audio content, thereby significantly improving situational awareness and response times.
Despite their advantages, most conventional SELD implementations are computationally intensive and require significant processing power, which makes real-time implementation on wearable or embedded devices challenging. In this work, we propose an efficient, real-time SELD system designed specifically for low-cost embedded hardware. Our approach focuses on optimizing both detection accuracy and computational efficiency to enable practical deployment on resource-constrained devices.
When a critical sound is detected, the SELD system extracts the relevant audio segment, processes it using HRTF to preserve spatial information, and seamlessly integrates it into the users ongoing audio stream. This ensures that while ambient noise remains suppressed, essential environmental sounds are conveyed with natural spatial perception. By subtly embedding these alerts within the users audio experience, our approach provides a non-intrusive yet effective method for improving situational awareness.
The ability to balance audio immersion and environmental awareness is crucial, particularly for users in urban environments, cyclists, or individuals who rely on ANC-equipped wearables in high-risk settings. Our proposed SELD system offers a practical and efficient solution that enhances safety without compromising the listening experience, paving the way for more intelligent and responsive wearable audio technologies.
(See document for exact affiliation information.)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
