Abstract
Substation safety notification (SSN) refers to the procedure that a project supervisor reads out the safety notification to the rest of his team, before carrying out any operation in a substation, to prevent any possible operation risks. Therefore, it is of great importance to make sure that the project supervisor has actually read out the safety notification sentence by sentence. However, at present, it relies on manual supervision, the workload of which is huge while the efficiency is low. Moreover, manual supervision can only be conducted afterwards, which cannot effectively reduce the phenomenon of skipping steps. In order to tackle this problem, this paper proposes to use artificial intelligence speech recognition to assist SSN on-site and in real time. The speech recognition algorithm proposed in this paper is based on improved parameter adaptive spectral subtraction (IPASS) and long short-term memory (LSTM) network. The algorithm first uses IPASS to preprocess the speech data of reading out the safety notification collected on the operation site, and then uses LSTM for speech recognition. Experimental results show that the proposed speech recognition algorithm has good anti-noise performance and can accurately recognize the on-site speech, and afterward, the speech will be matched with the safety notification to check if it has been read properly.