Multi-Head Attention Tacotron: A Chinese Speech Synthesis Model Based on Neural Network

Speech and text are one of the most important ways for human beings to receive and express information, and they play a vital role in people's lives. Speech synthesis is a technology that converts text information into speech signals. In recent years, with the improvement of computer computing power and the development of deep learning, research in the field of speech synthesis has made rapid progress, and many novel and efficient synthesis methods have emerged, among which the method based on end-to-end model has received extensive attention. The quality of speech greatly affects human understanding of text, and the quality of the current end-to-end model synthesis speech needs to be improved, and its attention mechanism and vocoder part need to be further improved. Therefore, to solve the above problems, this paper conducts an in-depth study of the existing speech synthesis systems and proposes a new model, Multi-Head Attention Tacotron. Based on Tacotron2, the model is trained on a Chinese dataset through the improvement of its attention mechanism and vocoder, aiming to improve the quality of Chinese speech synthesis. Eventually, the MOS value of synthesized speech was increased from 3.82 to 4.12, proving that the method we used was effective.