Smartphones with integrated sensors have really helped drive down the cost and power consumption of sensor-based solutions. Lower cost and lower power sensor-based solutions have, in turn, enabled the “Internet of Things” (IoT) market to become established. However, IoT applications are different than smartphone applications in many ways.
One of the most important is how power is managed by the user. In the case of smartphones, the user expects that their device will need to be recharged on a regular basis (often daily) and behaves accordingly. In the case of IoT solutions, however, the user expects that their device (or devices) will operate for extended periods of time (months or years) before needing to be recharged. Thus designers of IoT solutions have particularly challenging power consumption targets to meet.
One vision for IoT is a sensor/actuator enabled environment that responds to your needs when you are present, and acts on your behalf when you are not present. Voice is a natural way to express our desires when we are present, so voice will be an important part of implementing this vision.
Delivering voice capabilities when connected to the cloud has never been easier, but the power required for communication makes cloud voice solutions unsuitable for the ultra-low power requirements of most IoT solutions. Local voice processing, also known as embedded voice processing, has significantly better power characteristics, but even so still not low enough for most IoT solutions.
For example, conventional MCU-based voice processing results in power consumption levels that can be as much as five times the levels needed to operate from a pair of AA batteries for a year. The solution to this power consumption issue lies in a sophisticated mixture of computational hierarchy, context sensitivity and a mixture of software and hardware co-processing.
At QuickLogic we’ve focused our energy on developing just such “holistic” system-level approaches to dramatically reduce power, starting with ultra-low power sensor processing SoC platforms. In future blog articles we’ll discuss some of the specific additional steps needed to create IoT applications which can run for a year or longer on one or two standard AA batteries. Stay tuned…
2 thoughts on “Battery-Powered Voice Processing for IoT Applications”
Okay I feel very curious about this now. Do you literally mean there are envisioned IOT devices that would run on two standard AA power cells and also benefit from responding to voice commands for a whole year before changing the battery? Those would be big and heavy relative to a wearable. I ask because so many devices of a form factor large enough to accommodate two AA power cells (like a music playing Amazon Alexa device for example) can benefit from responding to voice commands, but do not run on two AA power cells. They require a greater capacity rechargeable battery or AC power source to support other things they do like playing music. Are you looking at applications like a module that does nothing but the voice command processing and then communicating the request wirelessly to a controller hub? That would be a device that can normally be be stationary, scattered around, say 6 or 8 of them in a home for example. It could enable voice commands from every room. If it sits on a table or shelf, and it does nothing but the voice command processing then I get why two AA cells would be okay for size and weight and also get why those lasting a year between changes would be nice. I hope you can give some indication what sort of applications you are referring to here.