It's an interesting idea, but I don't know enough about the internals of the recognition engine we're using to know if it's trainable to any sound, or if it's too specialised to human voices. I know it does quite a bit of preprocessing...

Hugo