Rionet 2In 2017, Rion developed and launched a series of hearing aids that incorporate the latest digital signal processing technologies to achieve sound reproduction faithful to the original sound. This series was positioned as the flagship model of Rionet hear-ing aids.Around the same time, Rion was also pursuing research on another technology. This research focused on AI-based noise reduction, which was conceived not necessarily as something for immediate commercialization, but pursued as a potential core technology for the future.The environment in which we live contains various types of noise. The hearing aids in the Rionet series incorporate a feature called NR (Noise Reduction) to suppress such noise. Noise reduc-tion can suppress stationary background noise, like noise from air conditioners or from ventilation fans present constantly over a certain period of time. But this NR method proved ineffective in suppressing transient sounds like the clattering of dishes or the rustling of a plastic bag. Could AI technology somehow be applied to solve this problem? Rion’s development team explored this possibility.Hearing aids incorporate a dedicated sound processor called a digital signal processor, or DSP, which converts sounds picked up by the hearing aid microphone into digital signals, then pro-cesses them into clearer, more distinct sound. Essentially, a DSP serves as the brain of a hearing aid.By 2020, prospects for a DSP for the next-generation prod-uct (Rionet 2) offering enhanced performance and capable of executing AI calculations efficiently were becoming plausible. This development made it increasingly realistic to consider incorporating AI technology, whose potential to become a core element had been explored, into hearing aids. The development team embarked on a project to create AI NR* (AI-powered noise reduction) that would suppress all background noise and deliver only the human voice to the ear.One member of the development team was Ryo Sato. To make a hearing aid capable of handling all types of noise, he drew on the diverse recorded data owned by the Environmental Instrument Division. Through near-endless trial and error, he performed numerous simulations to create an AI that would suppress noise effectively and achieve high noise reduction per-formance in various sound environments. However, it remained to be seen whether the AI calculations could be executed in real time inside a hearing aid.“Initially, I had no clear idea whether the AI calculations could be performed in real time by the DSP, as in the simula-tions,” confesses Sato. “I wasn’t optimistic. It wasn’t until we confirmed that the calculations for our AI would actually run in real time on the new DSP that I thought it just might work.”From the moment Sato saw this faint light at the end of the tunnel, the research project gained significant momentum.Conventional NR is a technology for suppressing stationary background noise constantly present over a period of time. The AI NR differs fundamentally in that it is capable of distinguish-ing between the human voice and other sounds and suppress-ing only the non-voice components. To achieve this, audio is analyzed approximately 25,000 times per minute, with each frequency band processed individually.Here, the greatest advantage held by Rion over its competitors was its vast database of noises and human voices. Development moved forward, using this database to simulate various sound scenarios.Research on AI-powered noise reduction technologiesMassive Volumes of Accumulated Noise DataThe two obstacles to integrating AI into hearing aidsArata YamadaHearing Aid Development Section, Development Department, Medical Instrument Division. Joined Rion in 2013. He has been involved in a wide range of activities from electrical designs for hearing aids to product development management. He has been assigned to the development of high-output hearing aids, the Rionet series, and Rionet 2. He is currently working on the development of new hearing aids.Ryo SatoAI R&D Group, R&D Department, R&D Center. Joined Rion in 2020. After working on signal processing and firmware development for hearing aids, he was assigned to the development of the AI NR function for Rionet 2. He is currently engaged in research and development for hearing aid processing methods based on deep neural networks.Rionet 2 is the first series with AI-equipped models. The enhanced digital functions achieve clear, natural, and audible sound. The rechargeable unit eliminates the need for battery replacements. On the left is the HI-C7 custom-fit in-the-ear hearing aid, made to fit the ear shape and hearing needs of individual users. It offers a perfect fit, feel, and comfort. On the right is the HB-A8 behind-the-ear (BTE) hearing aid, which can accommodate a broad range of hearing loss, from mild to severe. The HB-A8 BTE hearing aid is available in four colors.Yuuki YunoSoftware Development Section, Develop-ment Department, Medical Instrument Division. Joined Rion in 2018. She has been assigned to the research and development of hearing aid firmware and signal processing. With Rionet 2, she has primarily taken part in the research and development of the feedback cancellation. She is currently working on developing peripheral functions that will enhance the usability of hearing aids.3* The AI NR incorporates the results of joint research with Professor Nobutaka Ono of Tokyo Metropolitan University.
元のページ ../index.html#5