An Emotion Speech Synthesis Method Based on VITS

People and things can be connected through the Internet of Things (IoT), speech synthesis is one key technologies. At this stage, end-to-end systems are capable synthesizing relatively realistic human voices, but current commonly used parallel text-to-speech suffers from loss useful information during two-stage delivery process, control features synthesized monotonous, with insufficient expression features, including emotion, leading to emotional becoming a challenging task. In paper, we propose new system named Emo-VITS, which based on highly expressive module VITS, realize emotion synthesis. We designed network extract global local reference audio, then fused feature fusion attention mechanism, so as achieve more accurate comprehensive The experimental results show that Emo-VITS system’s error rate went up little bit compared without emotionality does not affect semantic understanding. However, superior other networks in naturalness, sound quality, similarity.