TSST

TSST数据集是由北京理工大学计算机学院的研究团队开发，专注于文本语音风格转换任务。该数据集旨在通过分析真实世界中的口语表达，提取出如情感表达、互动性、生动性和独特口语特征等多维度的口语风格特征，从而训练和评估大型语言模型在生成具有口语风格文本方面的能力。数据集的构建涉及从新闻、论文摘要和维基百科文章中筛选和处理数据，以确保多样性和适用性。TSST数据集的应用领域包括提升人机交互的自然性和效率，以及探索和增强语言模型在理解和模拟人类认知过程中的能力。

The TSST dataset was developed by a research team from the School of Computer Science and Technology, Beijing Institute of Technology, and focuses on the text speech style transfer task. This dataset is designed to extract multi-dimensional spoken style features including emotional expression, interactivity, vividness, and unique spoken characteristics by analyzing real-world oral utterances, so as to train and evaluate the performance of large language models (LLMs) in generating texts with natural spoken styles. The construction of the TSST dataset involves screening and preprocessing data sourced from news articles, academic paper abstracts, and Wikipedia articles to ensure data diversity and applicability. Application scenarios of the TSST dataset include improving the naturality and efficiency of human-computer interaction, as well as exploring and enhancing the ability of language models to understand and simulate human cognitive processes.