Speech-recognition on your tablet now possible

Leti recently helped create a demonstrator for a local, on-the-fly speech processing system using a multicore platform similar to those used in today’s tablets. The demonstrator has a 5,000-word vocabulary and can decode speech at speeds that set a new record for embedded applications. Currently, the majority of speech recognition systems use remote servers, which depend on fluctuating network performance.
Leti engineers are developing the technology in association with two manufacturers and a team of researchers at Inria. They are using middleware created by Leti to interface the application with sensors and actuators for home automation systems, voice-controlled roller shutters, HVAC systems, presence detectors, and more. The technology could ultimately help the disabled regain their independence at home. The research is part of the Rapsodie project, financed in part by the French Single Interministerial Fund.

Contact: diego.puschini@cea.fr

Facebook

Twitter