EXPRESSO
收藏资源简介:
EXPRESSO数据集由Meta AI创建,包含47小时的北美英语表达性语音,分为阅读和即兴对话两部分,涵盖26种自发表达风格。数据集通过专业录音室录制,质量高,格式多样。创建过程包括演员根据特定情境即兴对话,以及阅读特定文本。该数据集旨在推动无文本语音合成技术的发展,特别是在表达性语音合成方面,解决传统语音合成中表达性和自然性的不足。
The EXPRESSO dataset, developed by Meta AI, comprises 47 hours of expressive North American English speech, categorized into two modalities: read speech and impromptu conversational sessions, covering 26 distinct spontaneous expressive styles. Recorded in professional studios, the dataset features high audio quality and diverse file formats. The dataset’s construction entails actors performing impromptu dialogues tailored to specific contextual scenarios, as well as reading predefined textual content. This dataset is intended to drive the advancement of text-free speech synthesis technologies, with a specific focus on expressive speech synthesis, and to resolve the limitations of expressiveness and naturalness present in traditional speech synthesis systems.

- 1EXPRESSO: A Benchmark and Analysis of Discrete Expressive Speech ResynthesisMeta AI · 2023年



