I just uploaded a video of one of my previous projects where I deployed a speech recognition system (based on Julius decoder software) running on a Beaglebone Black computer, which runs a Debian-based OS. The video can be found in this project's “files” section The system was configured as a speech recognition server, which is accessed via Wi-Fi on a local network by an Android client application.
The difference to what I'm proposing now is that the speech recognition system will run on the Android-based smartphone (thanks to Google's native API); and the communication to the single board computer (now NTC's C.H.I.P.) that sends control commands to the TV will take place via Bluetooth rather than Wi-Fi.
This video is also available on YouTube :)