You are here

Embedded Distant Speech Recognition

The Shine unit is developing solutions for Embedded Distant Speech Recognition (EDSR). Small, cheap and low power hardware (built using digital MEMS microphones arrays and ARM single-board computers such as the RaspberryPi, SnowBall or Odroid) is combined with speech technologies in order to generate smart objects which are able to perform the following tasks:

  • keyword spotting
  • command recognition
  • speech transcription

The EDSR systems operate in real-time "always-listening" mode and are entirely self-contained within objects which ensures privacy as no audio is transmitted over the Internet. The prototypes developed by the Shine unit show the feasibility, potential and limits of the integration of speech technology in devices of everyday use.

The Shine unit was invited to the RAI TV show GeoScienza in April 2013 to present speech technologies applied to home automation and to demonstrate Embedded Distant Speech Recognition. The full episode can be seen in the following video.