Detection of missprounciation using deep learning
Varun Naga, Nandini K. Naga, M. Nandhakeerthi, M. Rupanjali · 2025
The voice recognition community is currently paying more attention to “mispronunciation detection”. This study interest and the focus of this work are primarily driven by two areas of application: speech recognition and language learning adaption. There are many Systems for CALL that use CAPT methods. This thesis introduces a new text-dependent mispronunciation method based on AFCC for text. This approach is demonstrated to perform better than the traditional HMM method based on MFCCs. To assist language learners in recognizing and correcting pronunciation errors, a PCA-based system for mispronunciation detection and classification is also developed. Two projects have been investigated in order to improve voice recognition through adaptation. As one method of making grammar-based name recognition adaptive, the initial one enhances name awareness by teaching permissible variances when pronouncing names. The second project involves detecting accents by looking at how fundamental vowels vary in speech with accents. It has been demonstrated that this method, which detects accents using both acoustic and phonetic information, is effective with accented English. To enhance name and speech recognition, these apps able to incorporated into a foreign phone system that is automated. Based on a short speech sample, it estimates the accent of the caller. For better detection outcomes, it moves away from the default switching from an accent-adaptive speech recognition engine once the type of accents is identified.