Sensory Announces Low-Power Voice Interaction Wake Words for Mobile Apps, Cuts Power Consumption by 80%
Sensory Inc. has announced that its TrulyHandsfree speech recognition software has been “reengineered to run ultra-low power on Android and iOS mobile applications,” without any special hardware requirements. TrulyHandsfree is positioned as the “first commercially successful always-listening low power wake word.” The product has been deployed on a wide range of device platforms such as Samsung, Huawei, LG, and Motorola as well as incorporated by chip manufacturers such as Qualcomm, Cirrus Logic and Conexant among others. The announcement goes into more detail on the wake word technology updates:
“In 2017 Sensory embarked on investigations of using Qualcomm and ARM as more standard cross-platform solutions to figure out how to lower power consumption for wake words used across mobile platforms. Sensory came up with a series of independent actions that when combined could lower power consumption on a mobile app using a wake word by more than 80%, or a reduction of approximately 200mAh in a 12-hour day. That enables a mobile app wake word to consume approximately one-percent of the smartphone battery in 12 hours. “
Little-Big Approach
The updates involve frame stacking and multi-threading techniques, but the more significant contributor to the improvement was what the company calls the “little-big” architecture. The wake word recognition service uses a small model that is always running to identify “interesting events,” such as the utterance of an invocation phrase. Power consumption of the small model is far less than the standard model, now referred to as the big model. The big model then becomes activated when the small model identifies an “interesting event” and ensures quality and accuracy are high for the user experience. The big model consumes more power, but it is activated less frequently which is critical in a mobile environment where battery life is finite.
While different from a technical implementation perspective and objective, the new TrulyHandsfree little-big approach is similar to how Amazon updated its Alexa wake word functionality in May 2017. Amazon had a number of challenges with false wake-ups because it was using a smaller model resident on an Echo device to determine if the wake word was spoken. The company updated its wake word approach last year to verify the wake word in the cloud where the service could employ a larger, more sophisticated model that tapped into cloud computing resources. This improved wake word accuracy which was the goal. In some ways that move by Amazon makes the Sensory announcement even more intriguing because TrulyHandsfree does all of this on a mobile device microprocessor. Sensory CEO Todd Mozer explained the new update this way:
“Hands-free operation for voice control has become the norm, and application developers are now looking to create hands-free wake words for their own apps. For example, we recently helped Google’s Waze accept hands-free voice commands by supplying them with Sensory’s ‘OK Waze’ wake word that runs when the app is open. With previous versions of TrulyHandsfree, having our always-on wake word engine listening for the OK Waze wake word during a short trip would have had minimal effect on a smartphone’s battery, but for longer trips a more efficient system was desired – so we created it.”
Even Runs in the Background on Android
One of the key benefits of Siri and Google Assistant on mobile is that their wake words are active all of the time. This has been a challenge for apps because wake words were only available when the app was active and in foreground. The TrulyHandsfree implementation can currently listen for wake words in background mode and activate the app when a user calls for it. A Sensory engineer commented:
The Low Power AP Wake-word solution can listen in background mode in the Android O.S. When the application running in the background hears the wake word it can move the application to the foreground for further user interaction.
He also added that the company is updating its iOS SDK this month to support similar low power performance for Apple. The foreground functionality should be the same as on Android, but iOS has some additional restrictions related to background listening. Final functionality scope for background listening on iOS will be dependent on the Apple certification process.
The key takeaway from this announcement is twofold. First, mobile apps now have a new tool to employ when looking to add voice interaction that includes always listening wake word functionality and also minimizes power consumption. Second, the availability of the wake word feature will be more accessible even when apps are in background. Most of the media and consumer focus is on speech recognition during the voice interaction. However, voice interactions start with a wake word and that is an entirely different set of technologies that provide the initial user access to hands free operation.
