Abstract
Speech, as the acoustic expression of language, is one of the most natural and effective means of human information communication. With the rapid development of the Internet and communication technology, the function of robot voice interaction is more and more popular among people. However, the robotic pronunciation cannot meet people's demand for personalized voice interaction. At the same time, style transfer technology, which is widely used in image and video processing, has been relatively mature. By studying the theoretical methods of generalized style transfer technology (including the style transfer of images and video signals), and comparing and analyzing various machine learning algorithms used by the current voice style transfer technology, this paper draws the following conclusions: First, various models generally have the problem of large demand for training data and difficulty in training. Second, an algorithm model shows the alienation effect in different usage scenarios. Finally, based on the above problems, suggestions for the development of voice style transfer are put forward.
Publisher
Darcy & Roy Press Co. Ltd.
Reference6 articles.
1. Ren Qiang. Research and application of speech style transfer technology based on Generative Adversarial Network [D]. Chongqing University of Technology,2019.
2. Gray R M. Vector Quantization[J]. Readings in Speech Recognition, 1990, 1(2):75–100.
3. Chu min, Lv Shinan. The invention relates to a synthesis method combining PSOLA algorithm and speech sine model [C]// Proceedings of the Fifth National Conference on Human-Computer Voice Communication. 1998.
4. Johnson J, Alahi A, Fei-Fei L. Perceptual losses for real-time style transfer and super-resolution[C]. European conference on computer vision, 2016: 694-711.
5. Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets[C]. Advances in neural information processing systems. 2014: 2672-2680.