Coco-Nut

Coco-Nut数据集是由东京大学开发的一个包含多样化日语语音样本的新型数据集，旨在推动基于自由形式文本描述的语音合成研究。该数据集包括高质量的语音数据、文本转录和自由形式的语音特征描述。创建过程涉及从互联网自动收集语音相关的音频数据、质量保证和众包手动标注。Coco-Nut数据集特别适用于训练对比语音-文本学习模型，以实现对语音特征的复杂控制，解决传统语音合成数据集覆盖范围有限的问题。

The Coco-Nut dataset is a novel dataset developed by The University of Tokyo containing diverse Japanese speech samples, aimed at advancing speech synthesis research based on free-form text descriptions. This dataset includes high-quality speech data, text transcriptions, and free-form speech feature descriptions. Its creation process involves automatically collecting speech-related audio data from the internet, performing quality assurance, and conducting crowdsourced manual annotations. The Coco-Nut dataset is particularly suitable for training contrastive speech-text learning models to achieve complex control over speech features, solving the problem of limited coverage in traditional speech synthesis datasets.