New Arm Hardware Could Make Voice Assistants 50 Times More Powerful
British tech firm Arm debuted a new microcontroller and neural processing unit for its AI platform that can make voice assistants perform as much as 50 times better than the current standard.
Faster, Smarter AI
The new Cortex-M55 microcontroller uses an advanced architecture that can make voice assistants understand and respond six times faster than they currently are able. When the new Ethos-U55 neural processing unit is used to augment the device, performance shoots up potentially 50 times the standard, while reducing power use by a factor of 25. That said, the theoretical maximum and the practical limit are not the same, so actual devices using the two new Arm creations will probably not reach those speeds. The improvement will still be notable for developers working on devices with the M55 and U55, which will come to the market in 2021.
Improving the AI’s efficiency and speed is very useful for Internet of Things devices. Cutting down the energy and processing needs opens up space for more sophisticated operations. While devices with the current Cortex-M chips are limited to vibration detection and similar alerts, the combination of the M55 and U55 means devices could measure and understand different voices and add speech recognition to gestures and visual cues to identify and distinguish users. And, because they are so efficient, the process can all be carried out on the device without sending information to the cloud.
“Enabling AI everywhere requires device makers and developers to deliver machine learning locally on billions, and ultimately trillions of devices,” Arm senior vice president and general manager for Internet of Things and Automobiles Dipti Vachani said in a statement. “With these additions to our AI platform, no device is left behind as on-device ML on the tiniest devices will be the new normal, unleashing the potential of AI securely across a vast range of life-changing applications.”
On the Edge of IoT
More powerful and efficient microcontrollers make putting technology ‘on edge,’ performing actions without using the cloud, much more feasible. The appeal to manufacturers and consumers of making smart devices that don’t need the cloud comes both from how they are less expensive and how they enhance privacy. Arm is already working with companies like Picovoice to provide platforms for voice assistants that don’t rely on sending data into the cloud. On-edge audio processing is part of the customizable voice assistant platform that speech tech developer Sensory debuted recently, for instance, while ID R&D came out with a new version of its vocal identification engine small enough to bring vocal security to devices without needing to transmit data. The ability to apply voice within a device is also what convinced Apple to spend a reported $200 million to acquire AI startup Xnor. Privacy and efficiency are strong incentives, and Arm is betting that that is where the future of smart devices lies.