Transducersaurus is a module which builds component WFSTsfor Automatic Speech Recognition Cascades (ASR). It containsclasses suitable for building language model transducers from ARPAformat LMs, lexicon transducers, context-dependency transducers and acousticmodel to context-dependency mappers.
It provides both simple python prototypes and more robust as well as much fasterC++ implementations of all the basic WFST algorithms needed to generate a standardLVCSR WFST cascade. It supports cascade generation for both the Juicer and TCubedWFST decoders. Sphinx support is on the way.