Phonetic Feature Recognition in Speech Training Application for Language Learners
Abstract
Language learners have a hard time saying words correctly because they don't know how to pronounce them and don't get quick feedback when they make mistakes. This study suggests a speech training app that utilizes phonetic feature recognition to assist individuals learning a second language in accurately pronouncing words. One way to achieve this is to utilize Mel-Frequency Cepstral Coefficients (MFCC) to train a Convolutional Neural Network (CNN) model on speech datasets with labels for classification and the extraction of phonetic features, such as voicing, place of articulation, and manner of articulation. The system instantly shows and hears feedback based on how well the user's input matches native phoneme models as they type. Key findings indicate that learners' articulation accuracy significantly improved, with phonetic feature recognition detection accuracy exceeding 85%, and learners' confidence increased substantially. The app can also help with personalized instruction by finding exact articulation errors. The results of this study suggest that phonetic-aware systems may be beneficial for helping individuals learn a language through personalized, interactive lessons.