A Method of Augmentation Mongolian Pronunciation Rhythm by Integrating Emotions

Aiming at the existing Mongolian speech corpus data scarcity and the lack of rhythmic representation and emotional expression, this paper proposes the Emotion-Integrated Pronunciation Prosody Control Model (EPCM), which realizes the emotional control of the generated pronunciation data through the introduction of fine-grained rhythmic features, and generates pronunciation data with different emotional expression and rhythmic expression. EPCM consists of a fine-grained emotion feature generation module, a text encoder, and a variable information adaptation unit, which can predict and adjust the subtle rhythmic features affecting the expression of emotions, such as pitch, duration, and intensity, to improve the natural fluency and pronunciation accuracy of the pronunciation data. After several sets of experiments, it was shown that the MOS of the generated pronunciation was 3.95, and the rhythmic expressions were 9.97%, 5.88%, 90.86% and 91.26% for GPDE, PDE, CSS and SS, respectively. The EMD was 89%.