Ave picoFlamingus. Julius te saluta!

Hi there!

This is our next step towards a fully portable picoFlamingo: Speech Recognition. This new feature makes the remote control device optional so, only the picoFlamingo box (will come at some point) is all you
need to make a presentation at any time in any place.

For speech recognition we are using Julius. A very nice tool that seems quite stable and powerful. Check out this video demonstration.

You should note that there is no microphone connected to the Zoom2. Well, we do not have a microphone to plug into the small jack connector in the Zoom2, so what we did was to use the remote audio broadcast tool adintool, distributed with Julius, to send the audio from out laptop to the Zoom2 through the network.

The command to do that is:

adintool -in mic -out adinnet -server server_hostname/IP

Then in the server you just need to specify adinet as input. Something like this:

julius -C ... -input adinnet

This speech control feature has been implemented as an add-on, in other words, as an special remote control client.

Basically we modified the julius-simple example in the Julius distribution to only dump to stdout the recognized command. Then we piped this commands to picoFlamingo using NetKitty. Quite simple.

For the acoustic model we got the one provided by voxforge. Actually there is a ready-to-use Julius version available there with all the data files you need to start working with the system.

The picoFlamingo Team


Nice one

Nice post