A Method for Evaluating Dubbing Quality Based on Multidimensional Acoustic Indicators and Emotional Dynamic Compensation and Its Application

Abstract

This application proposes a method for evaluating dubbing quality based on multi-dimensional acoustic indicators and dynamic emotional compensation, belonging to the field of intelligent speech and natural language processing technology. It aims to address the problems of subjective inefficiency in existing manual dubbing review and rigidity in automated linear scoring. The method includes: acquiring user dubbing audio, reference audio, and corresponding text; calculating pronunciation accuracy, intonation fit, and emotional expressiveness scores respectively; comparing the intonation fit with a preset hard threshold; if it is lower than the threshold, terminating the emotional compensation process and implementing a downgrade cap; if it meets the threshold and the emotional expressiveness satisfies preset trigger conditions, activating a non-linear emotional compensation mechanism to recover and compensate for pronunciation accuracy based on emotional expressiveness and intonation fit; finally, outputting the quality grading result based on the final score. This invention is mainly used for the automated objective evaluation and screening grading of massive amounts of dubbing works.

Images