Skip to the stories
Electronics & Semiconductors

Audio AI in Wearables: MEMS Microphones, Edge Processing and Context

  • Calin Miclaus
  • 656 reads

Low-power microphones and local AI let wearables respond to speech and environmental sounds while managing battery use, latency and privacy.

Audio AI in Wearables: MEMS Microphones, Edge Processing and Context

Wearable devices can be more useful when they respond to speech and environmental sounds without asking users to reach for a screen. Making that interaction practical depends on a chain of design choices: microphones must capture clean audio, low-power sensing must identify when processing is needed, and edge AI must interpret sounds quickly within tight limits on size and battery capacity.

This approach is relevant across smartwatches, earbuds, hearing devices, glasses, rings, pins and necklaces. Some newer form factors may have no display at all, making voice and sound awareness central to how users interact with them. Audio can support hands-free commands, spoken reminders and contextual assistance, while giving a device information about activity around its wearer.

Audio as an interface and a source of context

Speaking can be more convenient than typing or navigating menus, particularly when someone is occupied with another task. A wearable could record a shopping list during conversation and read it back later, for example. The value is not just command recognition: audio AI can combine what a person says with sounds in the surrounding environment to provide assistance that fits the situation.

Environmental sound recognition could help a device identify a siren, smoke alarm, breaking glass, approaching traffic or someone calling the wearer’s name. Those are potential applications, rather than performance claims for any particular product. Their usefulness will depend on reliable detection and timely alerts, without frequent false alarms.

Audio can also contribute to accessibility. The WeWalk Smart Cane 2 combines microphones with other sensor technologies to give users richer information about their surroundings. Its functions include letting users ask questions about nearby places, learn about landmarks and get everyday information. The example illustrates why combining sensing methods can provide more context than audio alone.

Figure 1: The WeWalk Smart Cane 2 combines microphones and other sensors to provide information about surroundings. Source: TDK InvenSense

Microphone design sets the limits

An AI model can only work with the signal it receives. Poor audio can reduce speech recognition accuracy and cause false activations, undermining the interaction even if the later processing is capable. Microphones therefore sit at the front end of the whole experience: their acoustic performance, power draw and physical dimensions affect what a wearable can do.

MEMS microphones suit wearable designs because they are small, light and require little power. Compact parts can leave more room for a battery, sensors or processing hardware, while giving industrial designers more freedom to make devices comfortable and visually unobtrusive. Those constraints are especially tight in products intended for all-day use.

Battery life is another practical limit. A device that listens continuously can drain its battery if the main processor and AI system must remain active at all times. TDK microphones include an acoustic activity detection (AAD) feature that monitors for relevant acoustic activity at lower power than keeping the primary processor awake. When the microphone detects an event, it can wake higher-level processing to analyze it.

This architecture aims to keep a wearable responsive without running its full processing stack continuously. The microphone handles the initial listening task; the application processor and AI inference are activated when needed. Local processing can also reduce the amount of audio sent elsewhere, which supports privacy and avoids unnecessary data transmission.

These benefits depend on where processing occurs and how the system is configured. AAD can screen for relevant sound before the main processor wakes, while edge AI can interpret audio on or near the device. Together, those functions can reduce response time and power use while keeping more interactions local.

Beyond speech and single-sensor input

Some MEMS microphones can detect ultrasonic frequencies, extending their sensing range beyond ordinary human hearing. That capability may support interaction or sensing methods that conventional speech-focused microphones cannot address. The specific applications depend on the device and its acoustic design.

Wearables can also combine microphone data with motion and magnetic sensors. Sensor fusion helps a system interpret what the wearer is doing and decide when to provide information, or when to stay quiet. For immersive audio, motion sensors can track head movement while microphones capture spatial sound, helping hearables and glasses maintain a three-dimensional audio field that responds to the wearer’s orientation.

Figure 2: Edge AI processes workloads on or near the wearable rather than sending every interaction to remote computing. Source: TDK InvenSense

Edge processing reduces delay and network dependence

After a wearable detects speech or another sound, it needs to interpret the input and choose a response. Edge AI runs that work locally or close to the device instead of sending every interaction to remote servers. Local processing can remove network delay, so commands such as setting a reminder or controlling music can be handled more quickly. It can also keep some functions available when connectivity is limited or absent.

Response speed matters to the feel of an interaction: a conversational interface becomes awkward if users have to wait after each request. Local inference also complements low-power listening by allowing the device to act on detected audio without transmitting every recording. The design challenge is fitting capture, detection and inference into a small device while meeting its battery and usability goals.

Smart glasses put the trade-offs in view

Smart glasses bring these constraints together. They need to remain light enough for extended wear and run on compact batteries while accommodating microphones, speakers, wireless connectivity, cameras, motion sensors and processors. Their familiarity may help: many people already wear prescription glasses or sunglasses, so the form factor can fit into daily routines.

But glasses don’t need to rely on graphics in the wearer’s field of view for every interaction. Audio, haptics, motion sensing and environmental awareness can deliver information with less visual interruption. Voice lets a wearer ask a question, capture a note, receive a reminder or request context without reaching for another device. Audio feedback can also keep assistance discreet.

Figure 3: Audio AI can support questions, note capture, reminders and contextual information in smart glasses. Source: TDK InvenSense

For this form factor, microphone power use and AAD affect how long the glasses can remain ready to respond. Edge processing can reduce dependence on connectivity and keep more interactions on the device. Those features help, but the experience still depends on reliable audio capture, useful sound detection and careful integration with the rest of the system.

What wearable designers need to get right

Audio AI’s progress in wearables rests on more than the capabilities of an AI model. It requires microphones that capture usable signals, low-power detection that avoids waking the main processor for every sound, and local processing that can respond within the device’s power and size limits. Motion and other sensors can add context that audio alone cannot supply.

The engineering measure is whether these pieces work together in an unobtrusive device that remains available through the day. MEMS microphones provide the audio input; AAD can limit how often higher-power processing runs; edge AI can interpret selected events locally. Their combination offers a route to faster, more context-aware interactions while managing battery use and personal audio data.

Source: Calin Miclaus · www.edn.com

From the same set