StoryTTS
收藏资源简介:
StoryTTS是由上海交通大学计算机科学与工程系创建的文本到语音合成数据集,包含从普通话讲故事节目中录制的61小时连续且富有韵律的语音,拥有精确的文本转录和丰富的文本表现力标注。数据集通过系统全面的标注框架,分析并定义了与语音相关的文本表现力的五个不同维度,包括修辞手法、句子模式、场景、模仿角色和情感色彩。StoryTTS旨在帮助未来的文本到语音合成研究充分挖掘丰富的内在文本和声学特征,特别适用于艺术作品的文本到语音合成研究。
StoryTTS is a text-to-speech synthesis dataset created by the Department of Computer Science and Engineering, Shanghai Jiao Tong University. It contains 61 hours of continuous and prosodically rich speech recorded from Mandarin storytelling programs, with precise text transcriptions and rich annotations of textual expressiveness. The dataset adopts a systematic and comprehensive annotation framework that analyzes and defines five distinct dimensions of speech-related textual expressiveness, namely rhetorical devices, sentence patterns, scenarios, imitated characters, and emotional tones. StoryTTS aims to facilitate future text-to-speech synthesis research to fully explore the abundant inherent textual and acoustic features, and is particularly applicable to text-to-speech synthesis research focused on artistic works.

- 1StoryTTS: A Highly Expressive Text-to-Speech Dataset with Rich Textual Expressiveness Annotations上海交通大学计算机科学与工程系 · 2024年



