This repository provides the official distribution of the AffetSpeech dataset. AffetSpeech is a large-scale emotional speech dataset with fine-grained, multi-level textual descriptions, enabling advanced research in speech emotion captioning (SEC) and emotional speech synthesis (ESS), containing 253,799 high-quality emotional speech data and 1,522,794 natural language descriptions.

• Emotions: angry, disgust, fear, happy, neutral, sad, surprise, contempt, calm
• Language: English
• Format: 16kHz, 16-bit, PCM wav.
• Annotations (ShareGPT format​): Includes sentiment polarity, open-vocabulary emotion captions, prosody (pitch, tempo, energy), emotional intensity, prominent segments, and semantic analysis.

AffetSpeech is released under a Restricted End User License Agreement (EULA) for non-commercial academic research purposes only.

1. Download the EULA.pdf from this repository.
2. Read and Sign: Please read the terms carefully. The form must be signed by a Permanent Staff/Faculty Member (e.g., Professor or Senior Researcher) of your institution.
3. Submit: Send the scanned PDF copy to qitianhua@seu.edu.cn.
   • Email Subject: [AffetSpeech Request] Name - Institution
   • Email Body: Please use your institutional email address (.edu, .ac, etc.) and briefly describe your intended use of the dataset.
4. Verification: Once approved, we will send you a private link to download the full version.

If you use AffectSpeech in your research, please cite:

@article{qi2026affectspeech,
  title        = {AffectSpeech: A Large-Scale Emotional Speech Dataset with Fine-Grained Textual Descriptions for Speech Emotion Captioning and Synthesis},
  author       = {Qi, Tianhua and Zheng, Wenming and Schuller, Bj{\"o}rn W. and Luo, Zhaojie and Li, Haizhou},
  journal={arXiv preprint arXiv:2604.04160},
  year         = {2026}
}

Qi, T., Zheng, W., Schuller, B. W., Luo, Z., & Li, H. (2026). AffectSpeech: A Large-Scale Emotional Speech Dataset with Fine-Grained Textual Descriptions for Speech Emotion Captioning and Synthesis. arXiv:2604.04160.

.
├── EULA.pdf               # License agreement to be signed
├── README.md              # Project description
├── metadata_sample/       # Small samples of annotations for preview
│   └── sample_sec.json
│   └── sample_ess.json
└── scripts/               # Helper scripts for data loading/evaluation
    └── data_loader.py