Text to Speech
Text-to-speech (TTS) research aims to synthesize natural-sounding human speech from textual input, focusing on improving speech quality, speaker similarity, and efficiency. Current efforts concentrate on developing advanced architectures like diffusion models and transformers, often incorporating techniques such as flow matching and semantic communication to enhance both the naturalness and expressiveness of generated speech. This field is crucial for applications ranging from assistive technologies and accessibility tools to combating deepfakes and creating more realistic synthetic datasets for training other AI models.
Papers
StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Models
Yinghao Aaron Li, Cong Han, Vinay S. Raghavan, Gavin Mischler, Nima Mesgarani
UniCATS: A Unified Context-Aware Text-to-Speech Framework with Contextual VQ-Diffusion and Vocoding
Chenpeng Du, Yiwei Guo, Feiyu Shen, Zhijun Liu, Zheng Liang, Xie Chen, Shuai Wang, Hui Zhang, Kai Yu