A team of MIT researchers have developed a speech recognition chip that they claim uses only 0.2 to 10 milliwatts of power for every full watt that a standard phone-based speech recognition system would use, reports MIT News. The technology is based on a neural network structure that only activates more complex sections of processing nodes when human speech is detected.
It’s a technology that could have important impacts in the emerging Internet of Things, where voice interaction technology is emerging as an important – and perhaps the dominant – user interface. While the system currently requires “a sizable onboard memory circuit for its intermediate computations,” as MIT News reports, it could find an increasingly wide range of applications if it is refined to perform on a smaller design scale – a project that could very well interest the MIT researchers partners at the Taiwan Semiconductor Manufacturing Company’s University Shuttle Program, which developed the chip prototype.
MIT isn’t unique in exploring this kind of cutting-edge speech recognition technology. Sensory, Inc. offers low-power speech recognition through its TrulyHandsfree solution, and major companies like Apple, Google, and Microsoft have been investing heavily in developing speech recognition technologies in their own platforms, recognizing its growing importance in products and services for consumers. But MIT’s researchers are clearly doing their part to push the state of the art further, which should inspire other organizations to stay competitive and result in more powerful technologies for end users down the line.
Source: MIT News
February 15, 2017 – by Alex Perala