Research on power system field speech recognition based on deep learning algorithm

This study focuses on the smart grid operation scenario, and explores the keyword extraction and analysis technology of the operators' oral operation process before admission. This paper proposes a speech recognition method based on hybrid architecture, which integrates the deep neural network hidden Markov model (DNN-HMM), Mel frequency cepstrum coefficient (MFCC), linear discriminant analysis (LDA) and other algorithms, and uses the weighted finite state transcription machine (WFST) and beam search algorithm to optimize the decoding and retrieval process. The experimental results show that the recognition algorithm in this paper is superior to the existing benchmark model in recognition accuracy and reasoning speed, and provides key technical support for improving the safety and efficiency of smart grid operation.