Year: 2026
Type: article
Abstract: To reduce word error rate of speech recognition in English interactive system, a key technology of speech recognition based on improved BiLSTM-HMM model is proposed.Firstly, BiLSTM-HMM speech recognition model is constructed by combining BiLSTM network and HMM model.Then, on the basis of BiLSTM-HMM model, attention mechanism is introduced, and model structure and data alignment are improved.Meanwhile, improved BiLSTM-HMM model is proposed, and it is applied to the English interactive system built on voice communication technology.Finally, verification is carried out on LibriSpeech dataset.The results shows that recognition time of improved BiLSTM-HMM model on LibriSpeech dataset is 3,541 s, and its error rate is 0.35; in English interactive system, error rate of English speech recognition for users is 0.34.Therefore, the key technology of speech recognition based on improved BiLSTM-HMM model can be used for speech recognition in English interactive system, and can effectively reduce the error rate of speech recognition in English interactive system, which has certain practical application value.
Authors: Xiaoling Ma, Qinmin Ma
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7166803136