Tuesday, January 31, 2006

Simple speech recognition

Suppose you want to assign names to commands for simple (look up the word) speech commands. You have a bunch of candidate spoken names for each command.

Consider a set of set of nodes, with a distance function defined between all pairs. Choose one exactly node from each set, maximizing the minimum distance between two chosen nodes, and also maximizing the second smallest distance after that, and so forth.

It becomes more interesting if arbitrary constraints across sets (these nodes must all be chosen together) are possible; multiple languages for the purpose of disambiguation.

No comments :