AI Speech Enhancement and the Future of Hearables
Abstract
Understanding speech in noise is the number 1 reported problem for hearing impaired people worldwide. Despite great signal processing advancements in the last 10 years, many people are still unsatisfied with how their hearables perform in this area. In our talk, we outline the speech enhancement problem for hearing aids and earbuds in noisy environments, explain how tight constraints limit what is possible, and show how our AI processors and speech enhancement algorithms are enabling large gains in background noise reduction, speech quality, and speech intelligibility. Finally, we look to the future of sensors in hearable devices, explaining how new sensors and algorithms are shaping the product definition of hearables and expanding the value offered to everyday users.
Biography
Scott Reid leads and oversees all AI model and toolchain R&D efforts at FemtoAI. After having studied Physics and Electrical Engineering at Stanford University and having worked in some of the most prestigious research labs in the country, Scott seized the opportunity to leverage his experience designing algorithms for Neuromorphic hardware to shape the direction of FemtoAI through developing their sparse AI technology. Scott has designed ultra-low power and latency AI speech enhancement and hearing aid algorithms, µW-scale spoken language understanding algorithms, and a flexible compiler and toolchain that optimizes virtually any AI model for tiny embedded deployment. Scott’s talents as a musician lend themselves to his unique approach to model development, blending rigorous signal processing techniques with artful inspiration from music and a keen ear for quality audio.