Web-based speech recognition chinese card game...

[imcgraw]

Hey all,

I posted a while ago about a free site that MIT is building for Mandarin studiers, but I want to give you an update as the features have changed dramatically.

http://web.sls.csail.mit.edu/chinesecards/?requireLogin=false

Now, in addition to building your own set of flash-cards with images in an easy Web 2.0 sorta way, you can load them up into a simple speech recognition enabled card game. There's even a 2-player mode where you and a friend can have a little "Word War". There's a demo video on the site if you don't have a microphone to test it out.

I think the best way to learn new vocabulary is to speak the new words in a meaningful context. That is a context in which the word is both understood and reacted to appropriately. This is really hard to do if there are no native speakers around... and if you're trying to learn a bunch of new words, very few native speakers will be patient enough help you through all of them.

This little card game won't get tired of you though! It works nicely for concrete nouns, descriptive adjectives, some simple verbs, etc... basically anything that can go in an image. And best of all: you decide the contents of the game.

Looking forward to your comments/feedback.

-Ian

[trevelyan]

What are you guys using for the backend speech recognition?

[imcgraw]

It's built by the Spoken Language Systems group at MIT, so it's they have their own in-house recognizer. The acoustic models are built from native speech data.

[imcgraw]

BTW, I'm really interested in how well the speech recognizer works for your voices... if you can write down your age and gender, and give me your thoughts on the recognition, I'd really appreciated it. Different voices will likely have very different experiences. Try the "animals" or "clothing" categories on the public set of cards for instance.

[trevelyan]

Is that open data/software, or is this a closed-source project?

[imcgraw]

To be honest, the project hasn't gotten to the stage where we've had to consider this decision yet. I don't believe MIT's speech recognition components have been made publicly available, but the rest of the components and the data could very well be open if there were enough interest. Obviously for parties connected to an established research institution, striking up a relationship with MIT's SLS lab would be easy.