Hugging Face's logo

Datasets:
ak0255
/
Synthesis_SER

ArXiv:
License:
Dataset card Files
xet
 Community
Synthesis_SER
File size: 2,934 Bytes
b5e203e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
---
title: Synthesis SER
emoji: 🎙️
colorFrom: purple
colorTo: blue
sdk: static
app_port: 7860
pinned: false
license: apache-2.0
---

# Synthesis SER Dataset

## Overview

This dataset contains **synthetic speech audio files** for Speech Emotion Recognition (SER) training, specifically designed for training the **emotion2vec** model. The data is generated using two state-of-the-art TTS systems: **CosyVoice2** and **Kimi-Audio**.

## Paper

If you use this dataset, please cite our paper:

```
https://arxiv.org/pdf/2603.16483
```

## Dataset Structure

Each subdirectory corresponds to a source emotional speech dataset and contains synthetic data generated using both TTS engines:

| Subdirectory | Source | TTS Engines | # WAV Files |
|-------------|--------|-------------|-------------|
| `CREMA_D_SYN.tar.gz` | CREMA-D | CosyVoice2 + Kimi-Audio | 2,236 |
| `IEMOCAP_SYN.tar.gz` | IEMOCAP | CosyVoice2 + Kimi-Audio | 3,570 |
| `LLM_DATA_SYN.tar.gz` | LLM_DATA | CosyVoice2 + Kimi-Audio | 3,155 |
| `RAVDESS_SYN.tar.gz` | RAVDESS | CosyVoice2 + Kimi-Audio | 12 |
| `SAVEE_SYN.tar.gz` | SAVEE | CosyVoice2 + Kimi-Audio | 124 |
| `TESS_SYN.tar.gz` | TESS | CosyVoice2 + Kimi-Audio | 12,397 |

### Internal Structure

Each tar.gz contains:
```
├── cosyvoice2/
│   ├── train/
│   │   ├── *.wav
│   │   └── train.json
│   └── test/
│       ├── *.wav
│       └── test.json
└── kimi-audio/
    ├── train/
    │   ├── *.wav
    │   └── train.json
    └── test/
        ├── *.wav
        └── test.json
```

Each `train.json` / `test.json` contains metadata:
```json
{
  "emotion": "angry",
  "wav_path": "..."
}
```

### Supported Emotions

- angry
- disgust
- fearful
- happy
- neutral
- sad
- surprised

## Real Data Sources

Since the real emotional speech data is publicly available, we only host the **synthetic** data on HuggingFace. If you need the **real** data for research, please download from the original sources:

| Dataset | Kaggle Link |
|---------|-------------|
| IEMOCAP | https://www.kaggle.com/datasets/samuelsamsudinng/iemocap-emotion-speech-database |
| RAVDESS | https://www.kaggle.com/datasets/uwrfkaggler/ravdess-emotional-speech-audio |
| RAVDESS_SONG | https://www.kaggle.com/datasets/uwrfkaggler/ravdess-emotional-song-audio |
| SAVEE | https://www.kaggle.com/datasets/ejlok1/surrey-audiovisual-expressed-emotion-savee |
| TESS | https://www.kaggle.com/datasets/ejlok1/toronto-emotional-speech-set-tess |
| MELD | https://www.kaggle.com/datasets/zaber666/meld-dataset |
| CREMA-D | https://www.kaggle.com/datasets/ejlok1/cremad |

## Usage

```python
from huggingface_hub import hf_hub_download

# Download and extract a dataset
repo_id = "ak0255/Synthesis_SER"
filename = "SAVEE_SYN.tar.gz"
path = hf_hub_download(repo_id=repo_id, filename=filename, repo_type="dataset")
```

## License

Apache 2.0