ATH-MaaS/CSEMOTIONS

--- task_categories: - text-to-speech library_name: datasets license: apache-2.0 language: - zh tags: - speech - emotional-speech - voice-cloning - mandarin dataset_info: features: - name: audio dtype: audio - name: text dtype: string - name: emotion dtype: string - name: speaker dtype: string splits: - name: train num_bytes: 3610108297.64 num_examples: 4160 download_size: 3077432286 dataset_size: 3610108297.64 configs: - config_name: default data_files: - split: train path: data/train-* --- # CSEMOTIONS: High-Quality Mandarin Emotional Speech Dataset [Paper](https://huggingface.co/papers/2508.02038) | [Code](https://github.com/AIDC-AI/Marco-Voice) [](LICENSE) **CSEMOTIONS** is a high-quality Mandarin emotional speech dataset designed for expressive speech synthesis, emotion recognition, and voice cloning research. The dataset contains studio-quality recordings from six professional voice actors across seven carefully curated emotional categories, supporting research in controllable and natural language speech generation. ## Dataset Summary - **Name:** CSEMOTIONS - **Total Duration:** ~10 hours - **Speakers:** 10 (5 male, 5 female) native Mandarin speakers, all professional voice actors - **Emotions:** Neutral, Happy, Angry, Sad, Surprise, Playfulness, Fearful - **Language:** Mandarin Chinese - **Sampling Rate:** 48kHz, 24-bit PCM - **Recording Setting:** Professional studio environment - **Evaluation Prompts:** 100 per emotion, in both English and Chinese ## Dataset Structure Each data sample includes: - **audio**: The speech waveform (48kHz, 24-bit, WAV) - **transcript**: The transcribed sentence in Mandarin - **emotion**: One of {Neutral, Happy, Angry, Sad, Surprise, Playfulness, Fearful} - **speaker_id**: An anonymized speaker identifier (e.g., `S01`) - **gender**: Male/Female - **prompt_id**: Unique identifier for each utterance ## Intended Uses CSEMOTIONS is intended for: - Expressive text-to-speech (TTS) and voice cloning systems - Speech emotion recognition (SER) research - Cross-lingual and cross-emotional synthesis experiments - Benchmarking emotion transfer or disentanglement models ## Dataset Details | Property | Value | |-------------------------|---------------------------------------| | Total audio hours | ~10 | | Number of speakers | 6 (3♂, 3♀, anonymized IDs) | | Emotions | Neutral, Happy, Angry, Sad, Surprise, Playfulness, Fearful | | Language | Mandarin Chinese | | Format | WAV, mono, 48kHz/24bit | | Studio quality | Yes | | Label | Duration | Sentences | | -------- | -------- | --------- | | Sad | 1.73h | 546 | | Angry | 1.43h | 769 | | Happy | 1.51h | 603 | | Surprise | 1.25h | 508 | | Fearful | 1.92h | 623 | | Playfulness | 1.23h | 621 | | Neutral | 1.14h | 490 | | **Total**| **10.24h**| **4160** | ## Download and Usage To use CSEMOTIONS with [🤗 Datasets](https://huggingface.co/docs/datasets): ```python from datasets import load_dataset dataset = load_dataset("AIDC-AI/CSEMOTIONS") ``` ## Acknowledgements We would like to thank our professional voice actors and the recording studio staff for their contributions. ## License The project is licensed under the Apache License 2.0 (http://www.apache.org/licenses/LICENSE-2.0, SPDX-License-identifier: Apache-2.0). ## 📜 Citation ```bibtex @misc{tian2025marcovoicetechnicalreport, title={Marco-Voice Technical Report}, author={Fengping Tian and Chenyang Lyu and Xuanfan Ni and Haoqin Sun and Qingjuan Li and Zhiqiang Qian and Haijun Li and Longyue Wang and Zhao Xu and Weihua Luo and Kaifu Zhang}, year={2025}, eprint={2508.02038}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2508.02038}, } ``` ## Disclaimer We used compliance checking algorithms during the training process, to ensure the compliance of the trained model and dataset to the best of our ability. Due to the complexity of the data and the diversity of language model usage scenarios, we cannot guarantee that the dataset is completely free of copyright issues or improper content. If you believe anything infringes on your rights or contains improper content, please contact us, and we will promptly address the matter. ------ 任务类别: - 文本转语音(Text-to-Speech, TTS) 库名称: datasets 许可证: Apache 2.0 语言: - 中文(zh) 标签: - 语音 - 情感语音 - 语音克隆(Voice Cloning) - 普通话(Mandarin) 数据集信息: 特征: - 名称: 音频 数据类型: 音频 - 名称: 文本 数据类型: 字符串 - 名称: 情感 数据类型: 字符串 - 名称: 说话人 数据类型: 字符串 划分: - 名称: 训练集 字节数: 3610108297.64 样本数: 4160 下载大小: 3077432286 数据集总大小: 3610108297.64 配置: - 配置名称: 默认 数据文件: - 划分: 训练集 路径: data/train-* --- # CSEMOTIONS：高质量普通话(Mandarin)情感语音数据集 [论文](https://huggingface.co/papers/2508.02038) | [代码](https://github.com/AIDC-AI/Marco-Voice) [](LICENSE) **CSEMOTIONS**是一款高质量普通话(Mandarin)情感语音数据集，专为表现力语音合成、情感识别及语音克隆(Voice Cloning)研究设计。本数据集包含来自6名专业配音演员的专业录音棚级音频，涵盖7种精心筛选的情感类别，可用于可控自然语言语音生成相关研究。 ## 数据集概览 - **名称：** CSEMOTIONS - **总时长：** 约10小时 - **说话人：** 10名（5男5女）母语为普通话(Mandarin)的专业配音演员 - **情感类别：** 中性、开心、愤怒、悲伤、惊讶、戏谑、恐惧 - **语言：** 普通话(Mandarin) - **采样率：** 48kHz，24位脉冲编码调制(Pulse Code Modulation, PCM) - **录制环境：** 专业录音棚 - **评估提示：** 每种情感对应100条提示，涵盖英文与中文 ## 数据集结构 每条数据样本包含： - **音频：** 语音波形（48kHz，24位，WAV(Waveform Audio File Format)格式） - **转录文本：** 普通话(Mandarin)转录语句 - **情感：** 属于{中性、开心、愤怒、悲伤、惊讶、戏谑、恐惧}之一 - **说话人ID：** 匿名化的说话人标识符（例如`S01`） - **性别：** 男/女 - **提示ID：** 每条语音的唯一标识符 ## 预期用途 CSEMOTIONS适用于： - 表现力文本转语音(Text-to-Speech, TTS)及语音克隆(Voice Cloning)系统 - 语音情感识别(Speech Emotion Recognition, SER)研究 - 跨语言及跨情感合成实验 - 情感迁移或解耦模型的基准测试 ## 数据集详情 | 属性 | 数值 | |-------------------------|---------------------------------------| | 总音频时长 | 约10小时 | | 说话人数量 | 6名（3名男性，3名女性，采用匿名ID） | | 情感类别 | 中性、开心、愤怒、悲伤、惊讶、戏谑、恐惧 | | 语言 | 普通话(Mandarin) | | 格式 | WAV(Waveform Audio File Format)、单声道、48kHz/24bit | | 录音质量 | 专业棚级 | | 情感类别 | 时长 | 语句数 | | -------- | -------- | --------- | | 悲伤 | 1.73小时 | 546 | | 愤怒 | 1.43小时 | 769 | | 开心 | 1.51小时 | 603 | | 惊讶 | 1.25小时 | 508 | | 恐惧 | 1.92小时 | 623 | | 戏谑 | 1.23小时 | 621 | | 中性 | 1.14小时 | 490 | | **总计**| **10.24小时**| **4160** | ## 下载与使用 若要结合[🤗 Datasets](https://huggingface.co/docs/datasets)使用CSEMOTIONS： python from datasets import load_dataset dataset = load_dataset("AIDC-AI/CSEMOTIONS") ## 致谢 感谢我们的专业配音演员及录音棚工作人员的贡献。 ## 许可证 本项目采用Apache许可证2.0（http://www.apache.org/licenses/LICENSE-2.0，SPDX许可证标识符：Apache-2.0）。 ## 📜 引用 bibtex @misc{tian2025marcovoicetechnicalreport, title={Marco-Voice Technical Report}, author={Fengping Tian and Chenyang Lyu and Xuanfan Ni and Haoqin Sun and Qingjuan Li and Zhiqiang Qian and Haijun Li and Longyue Wang and Zhao Xu and Weihua Luo and Kaifu Zhang}, year={2025}, eprint={2508.02038}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2508.02038}, } ## 免责声明 我们在数据集制作过程中使用了合规性检查算法，尽力确保训练模型及数据集的合规性。由于数据本身的复杂性及语言模型应用场景的多样性，我们无法保证本数据集完全不存在版权问题或不当内容。若您认为任何内容侵犯了您的权益或包含不当内容，请联系我们，我们将及时处理相关事宜。