Qualcomm is working on a speech and voice recognition system designed to run on-device, and the company’s researchers say its word and phrase recognition is 95 percent accurate, according to a VentureBeat report.
One of the company’s artificial intelligence researchers, Chris Lott, delved into the technology at the annual Re-Work Deep Learning Summit in Boston this week. The AI system is designed to run on a mobile device, and leverages recurrent neural network and convolutional neural network computing models. According to Lott, its machine learning technology can recognize the behavioral patterns of a particular user, and personalize its own functionality for that individual.
This kind of voice interaction technology has long been of interest to Qualcomm, with the company having partnered with Sensory Inc. back in early 2015 to integrate its low-power voice recognition solution into its platforms. Qualcomm is also increasingly interested in on-device AI, having announced a partnership with Baidu in late 2017 that would bring its DuerOS conversational AI technology into its Snapdragon mobile chip platform – a partnership that the companies announced this week would expand significantly as they seek to further leverage their combined AI capabilities.
Of course, processing voice-based queries in the cloud offers the salient benefit of internet access. But Lott tells VentureBeat that on-device processing will “make interacting with devices more natural,” and Qualcomm is evidently betting that this is the approach to invest in going forward.
May 29, 2018 – May 28, 2018