The 2015 Speech Industry Star Performers: Sensory
Sensory Squeezes Large Vocabularies Into Small Devices
Sensory is no newcomer to our Star Performer awards. The company received this honor last year for the introduction of its TrulySecure speaker verification technology, and for further developing its TrulyHandsfree Voice Control technology, which awakens consumer electronic devices once their name is called and enables them to respond without a button pressed. The innovation of the latter product has raised a few eyebrows, even landing Sensory partnerships with Samsung, Motorola, and Qaulcomm to incorporate the voice technology into their mobile devices.
While these deals are certainly worth celebrating, especially for a small company, Sensory isn't resting on its laurels. The company is innovating again and being recognized this year for the May release of its TrulyNatural solution, a client-based large-vocabulary speech recognition platform that facilitates natural language interactions. That the solution is client-based might not sound all that impressive, until you learn that the intended clients are small consumer electronic devices.
TrulyNatural can be used in smaller devices with less than 1 MB of memory that need vocabularies of only a few hundred words. It can also be used in smartphones, cars, and robots that require more natural language interfaces, capable of recognizing a million words and phrases. (In fact, Jibo, which makes family robots for the home, has already signed on as the first licensee of TrulyNatural.)
Sensory can deliver natural language interfaces to small devices because it uses a proprietary form of neural network with deep learning to achieve acoustic models significantly smaller than conventional solutions. In designing the technology, the real challenge was to make it robust enough while keeping it small enough to run on chips inside the most basic consumer electronics, said Todd Mozer, CEO of Sensory, during a press conference.
To deliver similar natural language interactions, other speech technology vendors have connected to large-vocabulary speech recognition platforms in the cloud. These systems generally require a lot of storage and processing power. While this approach has its benefits, there are some trade-offs. For example, wireless consumer devices might not always be able to connect to the Internet. And when they do, the data can be exposed to privacy and security threats.
Because TrulyNatural-equipped devices don't have to connect to the cloud, search queries using the solution can be processed much faster. This doesn't mean the solution can't connect to the cloud. It can, if needed.
The technology is available in U.S. English, Mandarin Chinese, Korean, Japanese, U.K. English, French, Spanish, and German. It will be available in Italian, Portuguese, and Russian later this year. SDKs are available for Android, iOS, Windows, Linux, and other platforms.
Additionally, the platform sets the stage for new developments. Later in May, for example, Sensory released its VoiceDial application, which adds hands-free voice dialing to Android smartphones and tablets. The app launches and places calls completely hands- and eyes-free. VoiceDial uses Sensory's TrulyHandsfree technology to listen for voice trigger phrases, and TrulyNatural technology to build and search the device’s contact lists by voice.
The user captures the application's attention by saying a trigger phrase. Then he can request a call to someone in his contact list. The application's speech recognition is speaker independent and does not require training. It can build and process large contact lists, and because the app processes the voice data on devices, it responds nearly instantaneously to call requests.
"VoiceDial…utilizes the robust capabilities of our recently announced TrulyNatural embedded large-vocabulary voice recognition solution. We designed the application to be simple in its function yet advanced in its performance, while being able to respond completely free of any haptic input from the user," said Bill Teasley, vice president of engineering at Sensory, in a statement.
Once again, Sensory refuses to be complacent. More innovation is in the works. "VoiceDial is just the tip of the iceberg," Teasley added. "We are very excited about the numerous other features our TrulyNatural technology will soon provide, such as seamless hands-free control of music, calendar, and navigation, just to name a few."