Image Number 2 for United States Patent #8117032.
A speech processing method including the step of identifying prerecorded audio comprising an original speech segment and a corresponding original noise segment. An audio stream can be generated from the prerecorded audio. The audio stream can comprise a stream speech segment and a stream noise segment. The stream speech segment can have approximately a same duration as the original speech segment. The stream noise segment can have a longer duration than the original noise segment. The audio stream can be conveyed to a speech recognition engine. The speech recognition engine can automatically determine an end of utterance condition based upon the stream noise segment. The original noise segment can be of insufficient duration for the speech recognition engine to determine the end of utterance condition. Responsive to the determining of the end of utterance condition, the stream speech segment can be speech recognized.