New opening for an Internship position: Weighted Finite State Transducer compression for Automatic Speech Recognition
WFSTs represent a means of pre-compiling the grammars used by ASR decoders. Such pre-compiled grammars tend to lead to both quicker decoding, because duplication of weights and paths is minimised, and simpler decoder implementations, because the composition process is offline. Pre-compiled grammars are quite large, sometimes limited only by the size of the memory on the host CPU.
More information about the position can be found on our