Magic Book -ASR Tech Demo

A.Russell · Post by **A.Russell** » Wed Feb 01, 2006 7:55 pm

We have just released a tech demo for testing ASR for game-based learning. It is based on Irrlicht and uses OpenGL, OpenAL and Microsoft SAPI.

This is a very simple game using automatic speech recognition. Try to read the highlighted sentences, either in one shot or in smaller chunks. The more words the computer hears, the higher your score! The better your pronunciation and fluency the more the computer will pick up.

While this was designed with game-based language applications in mind, it has also been suggested that this type of application could be useful for people with speech impediments.

This is freeware. We would really appreciate your feedback.

http://www.my-virtual-classroom.com/dow ... okDemo.msi

Or visit the homepage: http://www.my-virtual-classroom.com

dhenton9000 · Post by **dhenton9000** » Wed Feb 01, 2006 8:25 pm

Got the game to start, and the book showed, but it wouldn't continue.

A.Russell · Post by **A.Russell** » Thu Feb 02, 2006 2:15 am

Thanks,

Perhaps it couldn't recognise your voice. The first word "the" is a tricky one.

I forgot to mention, the X key is a cheat key. So if you really get stuck -especially the first word, just hit X.

Thanks for posting, I'll add that info to the ReadMe.

dhenton9000 · Post by **dhenton9000** » Thu Feb 02, 2006 2:42 am

It recognized "start the game" quite well, and could tell the difference between "start the banana" and "fritz the game" that was out of the box.

A.Russell · Post by **A.Russell** » Thu Feb 02, 2006 2:54 am

It depends quite a bit on how you have the microphone set up, too. If it is too sensitive it might just return lots of bad recognitions. You hear a "ding" sound when the computer thought it heard something, but couldn't recognise it. Sometimes it might get a recognition when you don7t say anything at all or mumble.

It is a good idea to go to the speech control panel (Start>>>Control Panel>>>Sound, Voice and Audio Device>>>Speech Recognistion (or something like that. I am on a machine with a Japanese OS, and that is my translation of what I see). You can set the microphone and its levels using the wizard. You can train it to understand you better, too, but then it adjusts to any bad speech habits you might have and sort of defeats the purpose.

It uses a dynamic command and control grammar to limit the number of words that the engine is listening for at any time. This improves recognition and makes it so you don't have to do the voice training.