Edge AI and Low‑Power Hardware Powering the Rise of Voice Control

Voice control is rapidly permeating virtually every consumer edge device. Thanks to breakthroughs in voice‑recognition algorithms and low‑power AI accelerators, even battery‑constrained appliances can now offer intelligent voice interfaces.

From the consumer perspective, the primary motivations for voice‑enabled smart home gadgets are straightforward.

“Ease of use and convenience drive adoption,” Alireza Kenarsari‑Anhari, CEO of PicoVoice, told EE Times. “Imagine commanding your office coffee maker from a desk or telling a tumble dryer what to do while you juggle laundry.”

Although many of these devices remain plugged in and always connected to home Wi‑Fi, processing audio locally offers compelling privacy and reliability benefits.

Edge AI is motivated by privacy—consumers want to keep their conversations local, and enterprises often require strict data controls—and by reliability. “Would you want your laundry machine to halt because of a Wi‑Fi outage?” Kenarsari‑Anhari asked.

Latency matters for time‑sensitive use cases, such as gaming, where even milliseconds can impact the experience.

Cloud inference incurs recurring costs per request, which is unsustainable for high‑frequency, low‑margin devices. Edge processing eliminates those charges.

PicoVoice offers a cloud‑free speech‑to‑text engine that runs on microcontrollers priced below $1. This opens voice control for wearables, hearables, and even industrial, security and medical devices that demand extreme power and cost efficiency.

Its new Shepherd platform lets developers create voice apps on microcontrollers without writing code, integrating with PicoVoice Console. Currently it supports popular ARM Cortex‑M MCUs from ST and NXP, with additional devices on the roadmap.

“I think of voice as an interface—if you can build your GUI or website without coding, maybe using WordPress, building voice interfaces in a similar way is the next logical step,” Kenarsari‑Anhari said. “Shepherd empowers product managers and UX designers to prototype and iterate quickly, and we aim to widen its target user base. What if everyone could build their own assistant? Name it whatever they want—no Alexa—and give it the personality they desire.”

While it is possible to develop natural language processing models from scratch, it requires significant resources. “Apple, Amazon, Google and Microsoft did it,” he added. “It’s a question of whether an enterprise has the capacity and willingness to build a dedicated organization around it.”

Voice is becoming the preferred interface for the next generation of technology users, Kurt Busch, CEO of Syntiant, told EE Times last summer.

Busch recalled his youngest child—who could read but not yet write or spell—using voice to text‑message friends on a smartphone. “Their generation got phones early; they now default to speaking,” he said. He envisions voice as “the touch screen of the future,” with in‑device processing delivering fast, responsive interfaces first on keyboards and mice, then on white goods.

Syntiant’s specialist AI accelerators handle voice workloads in consumer electronics with low power budgets. To date, the company has shipped over 10 million chips worldwide, most of which power mobile phones’ always‑on keyword detection. The latest NDP120 can recognize hot words such as “OK Google” and activate the assistant in under 280 µW.

Busch also sees voice as a democratizing force. “Three billion people live on $2 a day and often lack internet or formal education. Speech is the natural interface for them,” he explained. “Voice‑first applications can bring technology to underserved regions, both cost‑effectively and comfortably.”

Rapid growth can lead to fragmentation, said Vikram Shirastava, senior director of IoT at Knowles. “The market splits by speech engine, hardware platform, operating system, and acoustic environment—home, doorbell, car, etc.”

Knowles offers DSP‑based voice control solutions tailored to different verticals. By grouping markets with shared needs—such as home controls, TV soundbars, and remote controls—it can deliver optimized, “one‑level‑below‑turnkey” products. “We launch multiple releases that address specific fragmentation points, enabling us to serve the verticals we target,” Shirastava said.

Knowles’ latest AISonic Bluetooth Standard Solution is a development kit for voice recognition in Bluetooth‑connected devices like smart speakers, wearables, and in‑vehicle assistants. Built on the IA8201 dual‑core DSP silicon, it runs keyword spotting, source classification, beamforming, acoustic echo cancellation, and source direction estimation simultaneously, all within 50 mW. This efficiency stems from an instruction set with almost 400 custom audio and AI instructions, allowing lower clock frequencies and reduced power consumption.

Sugr’s iOttie Aivo Connect vehicle smartphone holder uses Knowles’ IA8201 for in‑car voice capabilities. It has Alexa voice assistant capability built‑in. (Source: Knowles)

Will voice become the default user interface across most consumer electronics? The convergence of advanced, efficient AI voice algorithms, developer‑friendly toolchains, and a growing ecosystem of low‑power, low‑cost hardware suggests it is headed that way.

>> This article was originally published on our sister site, EE Times.