English oral evaluation and pronunciation optimization algorithm based on speech recognition technology

The traditional methods of English oral evaluation and pronunciation optimization suffer from high subjectivity and delayed feedback. In response to this issue, this article aims to use speech recognition technology to provide objective and real-time oral evaluation and personalized pronunciation optimization guidance. Firstly, this article establishes a corpus containing rich phonetic data, including English oral samples with different accents and pronunciation characteristics. Secondly, this article adopts a DNN (deep neural network)-HMM (hidden Markov model) hybrid model for speech recognition, recognizing learners’ oral expressions and analyzing pronunciation errors. Then, personalized pronunciation optimization guidance is provided based on the recognition results. Finally, this article conducts experiments to verify the effectiveness of the algorithm. The experimental results show that the algorithm has made significant progress in spoken language assessment and pronunciation optimization, and the DNN-HMM model shows better performance in terms of the phonological error rate, word error rate and grammatical error rate, with the average phonological error rate, average word error rate and average grammatical error rate reaching 7.9%, 13.3% and 5.1% in multiple corpora, which significantly outperforms GMM (Gaussian mixture model)-HMM model, RNNs (recurrent neural networks), LSTM (Long Short-Term Memory), and Transformer. Moreover, participants demonstrated noticeable improvements in speaking speed, and their overall scores rose significantly, reaching a fluent level after utilizing the system for pronunciation training. The integration of DNN-HMM offers a novel approach for English speaking assessment. Through precise error detection and personalized optimization recommendations, learners’ pronunciation skills and overall speaking proficiency were significantly enhanced.