基于改进Transformer-TTS的情感语音合成方法研究

打开文本图片集

中图分类号：TN912 文献标识码：A 文章编号：1672-3791（2026）03-0050-03

Research on Emotional Speech Synthesis Method Based on Improved Transformer-TTS

JIA Ning ZHANG Tianmu

DalianNeusoft University of Information，Dalian，Liaoning Province，116o23 China

Abstract：This article proposes an emotional speech synthesis method based on improved Transformer-TTS， which introduces an emotional atention mechanism and integrates emotional embedding vectors to enhance the model'sunderstanding of emotional context，making the generated acoustic features more emotional.At the same time，a sentiment control module is designed to achieve precise control over the emotional intensity ofsynthesized speech byadjusting the weight ofthesentiment embedding vector.The experimental results show thatthe effectivenesof the emotional attention mechanism and emotional control module has been verified，and the emotional expresson of synthesized speech is more natural and authentic.

Keywords： Transformer-TTS modle; Emotional speech synthesis; Emotional attntion; Acoustic feature

随着人工智能技术的飞速发展，语音合成技术应用场景日益广泛[1]。（剩余3030字）