Domain Knowledge Enhanced Error Correction Service For Intelligent Speech Interaction

Intelligent speech interaction systems have gained great popularity in recent years. For these systems, the accuracy of automatic speech recognition (ASR) has become a key factor of determining user experience. Due to the influence of environmental noise and the diversity and complexity of natural language, the performance of ASR still cannot meet the requirements of real-world application scenarios. To improve the accuracy of ASR, in this paper, we propose a domain knowledge enhanced error correction method which first the improved phonetic editing distance to select the candidates which have the same or similar phonetics with the error segment, and then adopts language model the find the most appropriate one from the domain knowledge set as the final result. We also encapsulate the method as a service with the Flask + Gunicorn + Nginx framework to improve the high concurrency performance. Experimental results demonstrate that our proposed method outperforms the comparison methods over 48.4% in terms of accuracy and almost 20-40 times concurrency performance.