Research on Speech Recognition of Power Grid Dispatching Based on Big Data and Deep Learning
Hong Zhang, Lin Xiao, Yan Peng, Qianhong Xiao
Abstract
The rapid growth of the grid size leads to a saturated workload for a limited number of dispatchers, and when large incidents occur, the sudden and substantial increase in the number of instructions and messages makes it difficult for dispatchers to deal with them in a timely manner. This paper introduces Long-Short Term Memory (LSTM) and Conditional Random Field (CRF) LSTM-CRF for grid scheduling speech recognition. An acoustic lexicon of dispatching speech based on grid professional and localized vocabulary is constructed, and an acoustic model of grid dispatching speech recognition based on big data and LSTM-CRF is established. The inverse spectral mean-variance normalization method is used in feature calculation to reduce the influence of channel and noise and to improve the robustness of the acoustic model.