In this paper, based on the collection of English accent data, the Mel frequency cepstrum coefficient is used to obtain English accent signal characteristics. Then LSTM is combined with connected temporal classification network to establish the LSTM-CTC model applied to English accent recognition and error correction in colleges and universities, and the efficacy of the model is verified and scrutinized. Based on this, a DSP chip is incorporated as a foundation to develop an automated accent recognition and error rectification system for college English, and the accent recognition capabilities of the system are examined. The outcomes indicate that the phoneme error rate and word error rate of the LSTM – CTC model are 13.63% and 18.52% respectively and the system in this paper can recognize six different types of English accents. Combining deep learning with speech recognition technology can enhance the automatic recognition and error correction ability of English accent in colleges and universities, and help students better master different types of English accents.
