System and method of automatic speech recognition using parallel processing for weighted finite state transducer-based speech decoding

A computer-implemented system of speech recognition comprising:

• at least one acoustic signal receiving unit;
• at least one processor communicatively connected to the acoustic signal receiving unit;
• at least one memory communicatively coupled to the at least one processor;
• and
• a weighted finite state transducer (WFST) decoder communicatively coupled to the processor, and to:
• form a weighted finite state transducer (WFST) having nodes associated with states and interconnected by arcs, and to identify at least one word or word sequence hypothesis;
• identify multiple sub-graphs on the WFST, each sub-graph having the same arrangement of multiple states and at least one arc;
• and
• propagate tokens in parallel through the sub-graphs,
• wherein the WFST decoder is to perform non-parallel propagation of one or more tokens to or from one or more states external to all of the multiple sub-graphs on a frame of the WFST that were processed in parallel.

At least one non-transitory computer readable medium comprising a plurality of instructions that in response to being executed on a computing device, causes the computing device to:

• form a weighted finite state transducer (WFST) having nodes associated with states and interconnected by arcs, and to identify at least one word or word sequence hypothesis;
• identify multiple sub-graphs on the WFST, each sub-graph having the same arrangement of multiple states and at least one arc;
• and
• propagate tokens in parallel through the sub-graphs,
• wherein to propagate tokens in parallel comprises updating a score for a token by factoring a prior score, transition weight, and destination state acoustic score in parallel for multiple sub-graphs stored as active supertokens of a frame.