Research and Application of Convolutional Network Voice Based on Substation Inspection Scenario

With the rapid development of smart grids in China, more intelligent systems are put into the power grid, and new scene requirements are generated. Consequently, we would like to propose a speech recognition method, which is based on the improved convolutional network, and an order system of voice work is established for the substation inspection work order scenario, with three different speech characteristics and a hidden Markov model. A dataset of 4500 audio segments was established by different accents and environments in each region, then evaluating the feasibility of the system model. Finally, in the speech recognition experiment, the error rate of speech recognition words is controlled at 10%, and the sentence error rate is below 7%. Meanwhile, the speech input speed reaches 400 words per minute, and the response time of the recognition results is less than 200 ms. In voiceprint recognition, the recognition accuracy reaches 98% based on the voice characteristics of users, which verifies the rationality and feasibility of the proposed method compared with traditional CNN, and improves the work efficiency of inspection personnel.