A weight pushing algorithm for large vocabulary speech recognition

Weighted finite-state transducers provide a general framew ork for the representation of the components of speech recognit i n systems; language models, pronunciation dictionaries, co ntextdependent models, HMM-level acoustic models, and the output word or phone lattices can all be represented by weighted automata and transducers. In general, a representation is n ot unique and there may be different weighted transducers real izing the same mapping. In particular, even when they have exactly the same topology with the same input and output labels , two equivalent transducers may differ by the way the weights are distributed along each path. We present aweight pushingalgorithm that modifies the weights of a given weighted transducer in a way such that the transition probabilities form a stochastic distribution. This results in an equivalent transducer whose weight distributio n is more suitable for pruning and speech recognition. We demonstrate substantial improvements of the speed of our recogni ti n system in several tasks based on the use of this algorithm. We report a45% speedup at83% word accuracy with a simple single-pass40; 000-word vocabulary North American Business News (NAB) recognition system on the DARPA Eval ’95 test set. With the same technique, we report a 550% speedup at 88% word accuracy in rescoring NAB word lattices with more accurate 2nd-pass models. We finally report a 280% speedup at 68% word accuracy for100; 000 first name-last name pairs recognition.