Vocal Music Mispronunciation Identification and Correction Based on Artificial Intelligence Technology

In order to avoid blindness in vocal teaching, it is necessary to seek the correct method of vocalization that conforms to the principles of physiological science and to correct the wrong method of vocalization. In this paper, we describe the differences of vocal vocalizations by MFCC coefficients and pitch features, and adopt LAS model to extract high-level features in speech signals. And different vocalizations are explored in different acoustic feature parameter spaces. Deep learning and transcribed text techniques in artificial intelligence are utilized to detect and correct the errors identified by vocalizations. The results show that the absolute error of vocal error vocalization correction is between -1.99 and 7.15 Hz, with an accuracy rate of 97.1729%, and the error is controlled within 3%, which meets the needs of vocal teaching. The identification and correction method proposed in this paper is feasible and has positive significance for the informational reform of vocal music teaching.