ASR Post-Processing Correction Based On Biological Entity Context

Speech recognition technology is an important means in human-computer, but due to environmental noise, voice tone and other demographic factors; speech recognition accuracy is not high in the actual application environment. How to improve speech recognition accuracy must be resolved. Therefore, this article is will to optimize the result.In this paper, we introduce the method of natural language processing (NLP) into the test error correction which is recognized by speech recognition. We distinguish the context by instances which are identified as name entities, then get error detection and correction for post-proposing for ASR. In a small sample set, the paper’s correction algorithm is better than Wang Xingjian’s results which is based on word confusion network error correction, and improves the correct rate of 42.4% percentage. The main works of this paper are:1. Build a web crawler; get the original corpus from the relevant website. It provide corpus for the database which is establish by appropriate context environment.2. Build the context which is composition by similar conceptual in biomedical, by application of named entity recognition technology.3. Through the voice recognition technology research, we find the phoneme is a key factor in speech recognition, we added it into text correction after the speech recognition, by the method of NLP.4. Improved sentence similarity computation in this paper, it can get a better error correction effect in this framework.At last, design error correction algorithm appropriate for this framework. Compared to previous error correction effect, it gets a significant increase accuracy, which is proving that the error correction system has reference significance.