Improving the Multi-stack Decoding Algorithm in a Segment-Based Speech Recognizer