Optimizing Speech Recognition and Evaluation Models in English Listening Training Using Machine Learning Algorithms

Abstract

This study investigates the optimization of speech recognition models in the context of English listening training using machine learning algorithms. Through the development and evaluation of a hybrid deep learning architecture comprising convolutional and recurrent neural networks, we demonstrate the efficacy of our model in accurately transcribing English audio recordings. Statistical analyses reveal a low Word Error Rate (WER) of 12.5% and a high Sentence-Level Accuracy of 85.3%, indicative of the model's robust performance in capturing spoken language patterns and nuances. Hyperparameter optimization experiments yield optimal parameter values, while cross-validation analyses confirm the model's generalization capabilities across diverse linguistic contexts. Despite observed errors, our findings suggest promising avenues for future research and model refinement, including the integration of contextual information and adaptive learning strategies. This study contributes to the advancement of technology-driven approaches to language learning and pedagogy, paving the way for personalized and interactive English listening training experiences that empower learners to achieve fluency and proficiency in English comprehension.