We’ve all experienced the frustration of trying to have meaningful conversations amidst distracting background noise: whether meeting friends for dinner at a restaurant, making a connection at a cocktail party, or discussing work in an open-plan office. In loud environments, the human ear and brain often struggle to distinguish between multiple sound sources, making it difficult to focus on a particular conversation or dialogue. As people live longer, this critical skill increasingly atrophies due to the prevalence of hearing loss, often leading to debilitating social isolation.
Researchers at the University of Washington have discovered that AI can surpass human capabilities in pinpointing sound sources, effectively silencing specific areas. Within a 2-meter radius, this proprietary sound bubble technology enables discreet conversations to take place while significantly minimizing external noise and interference from other audio systems.
The research group, spearheaded by a professor at the University of Washington, aims to combine artificial intelligence with cutting-edge hardware to amplify human abilities. The challenge lies in developing practical AI applications within the boundaries of hardware limitations, particularly for mobile or wearable devices, as Gollakota notes, which diverges significantly from leveraging massive computational resources akin to those used today. Gollakota has long believed that the so-called “cocktail party problem” – where human ability to effortlessly converse with multiple individuals at once – can be tackled and leveraged through this approach.
Currently, commercial headsets effectively suppress ambient noise, but they do not account for varying distances to sound sources or the effects of reverberation in enclosed spaces. Previous studies have demonstrated that they achieve better source separation capabilities than traditional signal processing methods. Based on this groundbreaking discovery, Gollakota’s team created an integrated hardware-AI “hearable” system capable of accurately identifying sound sources within and outside a designated boundary without the need for additional dimensional constraints. As the system processes and synchronizes audiovisual data in real-time, it effectively eliminates distracting background noises, ensuring seamless synchronization of spoken words with visual cues, allowing for an uninterrupted viewing experience.
The audio component of the system features an industrial-grade noise-cancelling headset equipped with up to six high-fidelity microphones that simultaneously capture both nearby and distant ambient sounds, providing rich data for neural network analysis. Designed to optimize performance, customized-built networks utilize spatial awareness algorithms to accurately calculate the distance to sound sources and dynamically adjust their range to capture only those falling within a predetermined programmable bubble radius of 1 meter, 1.5 meters, or 2 meters. These networks have been trained with each simulated and real-world data, gathered in 22 rooms of varying sizes and sound-absorbing properties, featuring diverse combinations of human subjects. The algorithm operates on a compact embedded CPU, either the Orange Pi or a similar device, transmitting processed information back to the headphones within milliseconds, allowing for seamless synchronization between listening and visual experiences.
Listen to the stark contrast between dialogue with the noise-canceling headset activated versus deactivated. What are the implications of social media for mental health in youth? Allen Faculty/College of Washington
The algorithm on this prototype reduced the sound intensity outside the empty bubble by approximately 49 decibels, effectively diminishing it to a negligible level, roughly 0.001% of its original magnitude.The depth of water recorded contained within the bubble. Despite being deployed in novel acoustic settings and interacting with distinct customer bases, the system consistently performed well, accommodating up to two internal audio systems within the designated sphere, while effectively mitigating the impact of one or two external interference sources that were potentially louder. As the conversation unfolded, the group welcomed a fresh voice within its fold.
In settings where clarity and ease of verbal communication are paramount, such as loud environments, the primary purpose of customizable noise-canceling systems becomes increasingly significant. The perils of social isolation are starkly apparent, and an innovative solution focused on fostering in-person connections could prove invaluable in mitigating its effects. Dr. Gollakota suggests that the simple act of paying attention to one’s auditory and spatial awareness can be a valuable tool for individuals looking to improve their personal communication skills.
Sound-bubble knowledge may potentially be seamlessly integrated into hearing devices. Swiss hearing aid manufacturers, along with several others, have incorporated AI components into their earbuds and listening devices. Considering the potential of adapting the innovative sound-bubble technology for seamless integration into a discreet and comfortable hearing aid design. To seamlessly operate, the advanced hearing aids must synchronise wirelessly with each other, either via earbud integration or a discreet behind-the-ear design, while simultaneously providing uninterrupted all-day performance despite their compact power sources.
Gollakota is confident that this goal will be attainable. As he notes, we’re now witnessing the convergence of hardware advancements and sophisticated algorithms, ultimately paving the way for seamless AI augmentation. “This isn’t about AI replacing jobs, but rather leveraging its potential to have a positive impact through a seamless human-computer interaction.”
From Your Web site Articles
Associated Articles Across the Internet
