Frustrated with the battery life in your always-listening devices? Analog voice recognition is the solution to your preroll problem, and it’s a perfect example of the old adage: “everything old is new again.” Analog neural networks are the secret to more efficient ultra-low power sensors, and longer battery life.<\/p>\n\n\n\n
Analog neural networks are the secret to more efficient ultra-low power sensors, and longer battery life.<\/p><\/blockquote>\n\n\n\n
Microprocessors have gotten so small and cheap over the last 20 years that embedded device designers have gotten spoiled: their first impulse is to digitize all their IoT sensor input, and feed it to their algorithms. This “digitize everything” strategy works well enough when your device is plugged into a wall socket, but drains batteries too quickly on battery-powered IoT sensors and always-listening devices.
<\/p>\n\n\n\nembedded device designers have gotten spoiled: their first impulse is to digitize all their IoT sensor input<\/p><\/blockquote>\n\n\n\n
Think about it: smart speakers, voice-activated remote controls, home alarm sensors: all these edge sensors are always listening because they need to capture “preroll”–the crucial 5 seconds of audio BEFORE you say “Hey Alexa, or Hey Google”, which means they’re always draining their batteries.
Analog neuromorphic processing to the rescue! Instead of digitizing everything to listen for your wake word, analog neural networks, like the one in Aspinity’s RAMP chip<\/a>, can detect human voice BEFORE the data is digitized. Analog in-memory computing means your battery-draining digital processors can stay asleep while the analog neuromorphic chip listens for voice, or whatever sound you train the analog neural network to recognize. And Aspinity’s RAMP chip<\/a> can even capture and deliver preroll!<\/p>\n\n\n\n