NaturalVoices

NaturalVoices是由德克萨斯大学达拉斯分校电气与计算机工程系创建的大规模、自发、富有表现力和情感的语音数据集，包含超过3,800小时的语音数据，源自MSP-Podcast数据集中的原始播客。该数据集通过自动化处理流程提取语音中的丰富信息，如情感和信噪比，适用于语音合成和转换任务。创建过程中，利用了最新的深度学习技术，确保数据的多样性和高质量。NaturalVoices的应用领域广泛，旨在解决语音转换中自然性和表现力的问题，支持电影配音、智能对话系统等多种应用。

NaturalVoices is a large-scale, spontaneous, expressive and emotional speech dataset developed by the Department of Electrical and Computer Engineering at The University of Texas at Dallas. It encompasses over 3,800 hours of speech data derived from original podcasts in the MSP-Podcast dataset. Through automated processing pipelines, rich information including emotion and signal-to-noise ratio (SNR) is extracted from the speech, making the dataset suitable for speech synthesis and speech conversion tasks. During its creation, state-of-the-art deep learning technologies were employed to ensure the dataset's diversity and high quality. With a broad range of application scenarios, NaturalVoices aims to address the issues of naturalness and expressiveness in speech conversion, supporting various applications such as film dubbing and intelligent dialogue systems.