New Analog Solution Solves IoT Wake Word Battery Drain
Battery life can be a problem when mobile devices are always listening for wake-up commands. Even passive listening requires a continual trickle of power to test every noise for the right words. Intelligent sensor startup Aspinity debuted their own answer to the conundrum today. Devices built with the Reconfigurable Analog Modular Processor (RAMP) platform could last as much as ten times as long as the current standard before needing recharging or replacement batteries.
Analog and Digital
“There’s always a challenge in trading off battery power against performance and capability,” said Aspinity CEO Tom Doyle in an interview with Voicebot. “What we do that is different from everyone else is process sounds in the analog domain. Analog is scary for many, especially if you went to school in the last ten years.”
Devices that respond to audio commands have to always be listening for the right sound. That usually means constantly digitizing every noise in a way that can be analyzed by the software in a device or in the cloud. That takes energy and processing power for what is mostly instantly discarded information.
“The paradigm of the last 30 years has been digital only, but lately we’ve been seeing a move to what we call the edge,” Doyle said. RAMP can identify speech in analog form before waking up the rest [of the system]. It takes an order of magnitude less power. All that being said, analog is hard.”
RAMP flips the usual protocol to analyze sound before digitization. The lower energy and processing power costs of doing so extends how long a battery charge lasts.
“The real play is if we can be more efficient at the edge then we can send less data to the cloud,” Doyle said. “A lot of times, things go to the cloud because they have to because you can’t remove processing power from the cloud. We enable a new level of keeping things local. Everything is turned off until we detect an event, and not sent to the cloud unless it’s the right sound. If [the information] is not digitized, it’s not going to the cloud.”
The processor and platform are flexible in terms of where they could be integrated. Any smart speaker, wearable, or other Internet of Things (IoT) device with the right sensors would work. Doyle said the goal is for two AAA batteries to double their lifespan from six months to a year in devices using the RAMP platform.
“We want to be able to provide chips for someone like Apple or Amazon or Google or others to have the headset or watch last longer,” Doyle said. “We are working with partners and do expect products to come out next year. We’re targeting headsets and anything that is battery-powered, like remote controls and smart home devices.”
Out of Stealth
Pittsburgh-based Aspinity was founded in 2015 but has been in stealth mode up until now. The company went through the Techstars Alexa Accelerator program in 2017 and closed a $3 million seed round near the end of last year from investors including the Amazon Alexa Fund.
The low-power listening space is picking up more attention as the spread of wireless headsets like AirPods become more common. Earlier this month, Knowles unveiled the first Amazon-approved Alexa headset developer kit. To get Amazon’s approval, their system had to be efficient enough not to drain batteries while always listening for the wake command. Similar to Aspinity, Knowles puts its operations on the edge of the system. The difference is that Aspinity’s system is analog compared to Knowles’ digital operation.
“Everybody is trying to solve the longevity problem, but all of those are using digital data,” Doyle said. “Other folks are trying to solve the battery power problem, but few are leveraging analog as we are.”