espnet/jesus_dramas
收藏资源简介:
Jesus Dramas数据集是一个包含430种语言的宗教音频戏剧的集合,总计约640小时的音频。音频以16kHz单声道格式存储,每个音频戏剧可能包含多个男女声的说话者。该数据集可用于语言识别、口语建模或语音表示学习。原始音频来源自InspirationalFilms网站。数据集用于训练XEUS,一个支持4000多种语言的多语言语音编码器。
Jesus Dramas is a collection of religious audio dramas across 430 languages, totaling around 640 hours. These audio files can be used for language identification, spoken language modeling, or speech representation learning. The dataset includes raw unsegmented audio in a 16kHz single-channel format, with each audio drama potentially containing multiple speakers of both male and female voices. The data was sourced from InspirationalFilms. This dataset is used to train the multilingual speech encoder XEUS, which supports over 4000 languages. The dataset contains three fields: id, language, and audio. It is released under the CC BY-NC-SA 4.0 license, and users are required to cite the relevant paper and acknowledge the original creators of the data.
数据集概述
基本信息
- 数据集名称: Jesus Dramas
- 语言: 多语言(430种语言)
- 总时长: 约640小时
- 采样率: 16kHz
- 声道: 单声道
- 用途: 语言识别、口语语言建模、语音表示学习
数据结构
- 特征:
id: 字符串类型,表示话语IDlanguage: 字符串类型,表示语言名称audio: 音频数据,采样率为16kHz
数据集配置
- 配置名称: default
- 数据文件路径:
data/train-* - 训练集:
- 样本数: 423
- 数据大小: 54665637580字节
许可证
- 许可证类型: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 (CC BY-NC-SA 4.0)
引用
-
论文引用:
@misc{chen2024robustspeechrepresentationlearning, title={Towards Robust Speech Representation Learning for Thousands of Languages}, author={William Chen and Wangyou Zhang and Yifan Peng and Xinjian Li and Jinchuan Tian and Jiatong Shi and Xuankai Chang and Soumi Maiti and Karen Livescu and Shinji Watanabe}, year={2024}, eprint={2407.00837}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2407.00837}, }




