AI Hearing Aids: How Artificial Intelligence Is Used in Modern Devices

Artificial intelligence turns up in a lot of conversations these days, and hearing technology is part of that story. Several manufacturers now describe their devices as AI hearing aids, using machine learning to process sound. This article is a neutral, technical explainer of how artificial intelligence is used in modern hearing aids and which brands describe these features. It reports what the makers say and does not recommend or rank any product.

What "AI" Means Inside a Hearing Aid

In this context, artificial intelligence usually refers to a form of machine learning called a deep neural network, often shortened to DNN. Broadly, a deep neural network is a computer system trained on large numbers of sound samples so it can recognise patterns, such as telling speech apart from steady background noise. Manufacturers describe building these networks into the tiny chip inside a hearing aid so the analysis happens on the device itself. This is a description of how the electronics are designed, not a claim about results for any individual.

How the Technology Processes Sound

The common thread across AI hearing aids is sound classification. According to the manufacturers, a trained network sorts incoming sound into categories many times a second and handles speech-like sound differently from constant background sound. The networks are described as being trained on very large sound libraries. It is a fast, automatic sorting job happening inside the device, which is the engineering idea at the heart of these features.

Brands That Describe AI Features

A number of manufacturers, all available through audiology clinics, describe artificial intelligence or machine learning in their current ranges. The following are listed neutrally as examples, in no particular order, each with its official source. These are the manufacturers' own descriptions of how their features are designed to work; CQ Audiology has not independently tested these claims, and how well any feature performs varies from person to person:

• Phonak describes its Audeo Sphere range as using a dedicated DEEPSONIC chip and a Spheric Speech Clarity feature to separate speech from noise, alongside its AutoSense OS system that identifies the sound environment automatically.
• Oticon describes its devices as carrying an on-board deep neural network, which the company says is trained on millions of real-world sound scenes to analyse the full sound environment.
• Starkey describes its Edge AI range as built around a G2 Neuro Processor with a neural processing unit that performs deep neural network sound processing on the device.
• Signia describes its Integrated Xperience platform and RealTime Conversation Enhancement feature as analysing conversation dynamics to track multiple speakers, supported by a cloud-based deep neural network in the Signia Assistant.
• ReSound describes its Vivia range as using a dedicated deep neural network chip, trained on millions of spoken sentences, together with an Intelligent Focus feature.
• Widex describes its SoundSense Learn feature, part of its MySound tools, as machine learning that works through the app to calculate sound settings from user input.
• Bernafon describes its flagship Encanta range as using machine learning, which it calls Machine Learning 2.0, to power its Smart Noise Reduction, a system it says is trained on many sound samples to tell speech apart from background noise.

CQ Audiology supplies and fits hearing aids from all the major manufacturers, including the brands named in this article, so we have a commercial interest in the devices we fit. This article is general information, not a recommendation of any particular brand or device, and the brands named are examples rather than the only options available. Which device suits a person is worked out individually with a qualified audiologist. The exact feature names and technologies differ between companies and change over time, so a manufacturer's own site is the place to check for current detail.

On the Device and in the App

There are two broad places this processing can happen. Some of the work runs on the chip inside the aid itself, which the makers describe as on-device processing. Other features, such as some of the learning tools, connect through a smartphone app and, in some cases, a cloud service. Manufacturers describe both approaches across their ranges. This is simply how the different features are organised, and the mix varies from brand to brand.

A Field That Keeps Moving

Artificial intelligence in hearing technology is an area that manufacturers keep developing, with new chips and features announced regularly. Because the technology is individual and the settings are configured by an audiologist to suit a person's device, the specifics of any one setup are a matter for an appointment rather than an article. What stays constant is the underlying idea: trained networks sorting sound inside a very small device.

Understanding how AI is used in modern hearing aids helps make sense of a lot of the marketing language, showing the shared engineering idea beneath the different brand names. If you would like to understand the technology in a particular device, we are happy to walk you through it.

Sources: manufacturer official information from Phonak, Oticon, Starkey, Signia, ReSound, Widex and Bernafon (deep neural network and machine learning features), linked above.