Direct construction of compact context-dependency transducers from data

This paper describes a new method for building compact context-dependency transducers for finite-state transducer-based ASR decoders. Instead of the conventional phonetic decision tree growing followed by FST compilation, this approach incorporates the phonetic context splitting directly into the transducer construction. The objective function of the split optimization is augmented with a regularization term that measures the number of transducer states introduced by a split. We give results on a large spoken-query task for various n-phone orders and other phonetic features that show this method can greatly reduce the size of the resulting context-dependency transducer with no significant impact on recognition accuracy. This permits using context sizes and features that might otherwise be unmanageable.

• Slight modifications of decision tree construction may impact size of CC transducer.
• We construct the CC transducer directly from data instead of decision trees.
• Transducer size is controlled by incorporating the state count in the optimization.
• Substantial reduction in transducer size is achieved without impact on ASR accuracy.
• We analyze the construction for larger phonetic context and generalized features.