A fast segmental clustering approach to decision tree tying based acoustic modeling

A fast two level segmental clustering approach to decision tree based state tying is proposed for large vocabulary speech recognition. This approach extends the conventional segmental K-means approach to phonetic decision tree tying based acoustic modeling. It achieves high recognition performances while reducing the model training time from days to hours, compared to approaches based on incremental Baum-Welch training. Experimental results for this fast segmental clustering approach are presented for resource management and the Wall Street Journal tasks.