Real Time Speech Recognition Method for Online Complaints from Power Grid Customers Based on Improved Residual Network

Online complaints may be subject to various types of sound interference, such as noise and echo, which can affect the quality and clarity of the speech signal, thereby affecting the accuracy of speech recognition. To address this issue, this paper proposes an electric power customer online complaint realtime speech recognition method based on an improved residual network. The white noise and partial discharge pulse signals of the electric power customer online complaint realtime speech are extracted, and the deep residual network is improved for waveform reconstruction of the speech signal. Linear prediction parameters and linear prediction cepstral coefficients are calculated to complete the speech recognition. Experimental results show that the proposed method can effectively recognize electric power customer online complaint speech in different signal-to-noise ratio areas, and has good speech signal-to-noise ratio recognition and positioning performance. The recognition rate is over 96%, and the recognition time fluctuates between 5ms and 25s. Moreover, the proposed method can achieve higher speech signal-to-noise ratios.