ASR post-processing correction based on NER and pronunciation primitive

摘要

In dealing with robustness of specific areas,such as automatic speech recognition (ASR).this paper proposes some new ideas. The idea of using named entity recognition(NER), which is domain-specific is based on the conditional random field(CRF).NE are used to establish the context, leading the speech recognition process' pronunciation element into the post-treatment of speech recognition, Speech recognition results are represented with pronunciation primitive characters. And based on the improved dynamic edit distance we find the appropriate entity context, and then according to the context of the entity we try to optimize the recognition results.