Applications

DTW pattern-matching may be used for speech recognition applications with a small vocabulary of stored "prototype" pronunciations. For example, given the task of recognising which of ten digits (0-9) was spoken, a spoken input can be compared to a library of ten recordings, and the closest-matching word identified. Other words than digits, in any language, could be used instead, without altering the method.

Exercise. Think of as many pitfalls for this method as you can. For example, will it work with a large vocabulary, or if it is used by someone with a heavy foreign accent? What would happen if you coughed into it?

Next: Vector quantization