A Study of Phonetic Interaction Quality in English Based on Deep Learning

To address the challenges of untimely correction of pronunciation errors and speech noise present in traditional methods for assessing English pronunciation interaction quality, we propose a novel deep learning-based approach. This method utilizes a deep learning-based natural language description model to design a lightweight algorithm for generating natural language descriptions. Building on this foundation, we design an English pronunciation interactive acquisition flow and preprocess the interactive process. An interference factor filter is introduced to reduce speech noise during English pronunciation. Additionally, an error detection model, grounded in Stochastic Forest Theory, is developed to identify errors in English pronunciation interactions. The English pronunciation corpus is collected and used for training, with relevant features extracted to evaluate the quality of the pronunciation interaction. The effectiveness of the proposed method is validated using a voice-sensing unit and an English speech interaction processor. Simulation results demonstrate that the proposed noise filtering method is effective, offering superior accuracy and quality in cross-checking English pronunciation compared to existing techniques.