As the years go by, I get lazier it seems. So the prospect of talking to my computing devices gets better and better as time passes (even if I have to yell across my house to my Google Home, which is amusing in its own right). However, always-on speech recognition comes at a price for battery-operated devices. Researchers at MIT claim to have to come up with a solution to this: a dedicated speech recognition chip that can reduce power consumption by 90-99% across real-world devices.
That is no small claim. The example given by the researchers is "a cellphone" that is actively listening for voice commands may consume about 1 watt of power, whereas this new chip may consume anywhere from .02 to .1 milliwatts, depending on how much it has to listen to.
Here is what Professor Anantha Chandrakasan, head of the project, has to say on it:
"Speech input will become a natural interface for many wearable applications and intelligent devices. The miniaturization of these devices will require a different interface than touch or keyboard. It will be critical to embed the speech functionality locally to save system energy consumption compared to performing this operation in the cloud."
Not only could we see this greatly impact the normal devices like smartphones, but it could be branched out into other, smaller electronics. With the chip, those platforms could run on a single charge for months or easily replenish energy from the surrounding environment.
So how does this work? As I am sure you all know, current devices with an always-on speech recognition feature are constantly listening for a command (go figure). This can be a huge drain on a battery, even if you do something like the original Moto X did with the custom X8 SoC — which had a sort of co-processor that exclusively handled language interpretation, offloading the responsibility from the CPU. MIT's chip will instead monitor the noise surrounding it, filtering out ambient sound from actual voices. If it does recognize human speech, it will fire up the larger, more complex recognition circuit to process any following commands.
This could be a big deal for future IoT and wearable devices that have small battery cells and yet are still required to actively listen for speech. If you're interested in more in-depth information, hit up the source link below.