File size: 4,345 Bytes
5d8fe69
 
95249d6
 
 
 
 
 
 
 
 
 
 
5d8fe69
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: gpl-3.0
language:
- en
tags:
- emotion
- speech
- facial
- text
- semantic
- multimodel
size_categories:
- 1K<n<10K
---

# Hi, I’m Seniru Epasinghe 👋

I’m an AI undergraduate and an AI enthusiast, working on machine learning projects and open-source contributions.  
I enjoy exploring AI pipelines, natural language processing, and building tools that make development easier.

## 🌐 Connect with me

[![Hugging Face](https://img.shields.io/badge/Hugging%20Face-seniruk-orange?logo=huggingface&logoColor=white)](https://huggingface.co/seniruk) &nbsp;&nbsp;
[![Medium](https://img.shields.io/badge/Medium-seniruk_epasinghe-black?logo=medium&logoColor=white)](https://medium.com/@senirukepasinghe) &nbsp;&nbsp;
[![LinkedIn](https://img.shields.io/badge/LinkedIn-seniru_epasinghe-blue?logo=linkedin&logoColor=white)](https://www.linkedin.com/in/seniru-epasinghe-b34b86232/) &nbsp;&nbsp;
[![GitHub](https://img.shields.io/badge/GitHub-seth2k2-181717?logo=github&logoColor=white)](https://github.com/seth2k2)
---

# Multimodal Emotion Recognition Dataset (Processed from MELD)

This dataset is a **preprocessed and balanced version** of the [MELD Dataset](https://www.kaggle.com/datasets/zaber666/meld-dataset), designed for **multimodal emotion recognition research**.  
It combines **text, audio, and video modalities**, each represented by a set of **emotion probability distributions** predicted by pretrained or custom-trained models.


## Overview

| Feature | Description |
|----------|--------------|
| **Total Samples** | 4,000 utterances |
| **Modalities** | Text, Audio, Video |
| **Balanced Emotions** | Each emotion class is approximately balanced |
| **Cleaned Samples** | Videos with unclear or no facial detection removed |
| **Emotion Labels** | `['angry', 'disgust', 'fear', 'happy', 'neutral', 'sad', 'surprise']` |

Each row in the dataset corresponds to a single utterance, along with emotion label, file name, and predicted emotion probabilities per modality.


## Example Entry

| Utterance | Emotion | File_Name | MultiModel Predictions |
|------------|----------|------------|----------------|
| You are going to a clinic! | disgust | dia127_utt3.mp4 | {"video": [0.7739, 0.0, 0.0, 0.0783, 0.1217, 0.0174, 0.0087], "audio": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0], "text": [0.0005, 0.0, 0.0, 0.0007, 0.998, 0.0004, 0.0004]} |

### Column Description:
- **Utterance** — spoken text in the conversation.  
- **Emotion** — gold-standard emotion label.  
- **File_Name** — corresponding video file (utterance-level).  
- **MultiModel Predictions** — JSON object containing model-predicted emotion probability vectors for each modality.


## Modality Emotion Extraction

Each modality’s emotion vector was generated independently using specialized models:

| Modality | Model / Method | Description |
|-----------|----------------|--------------|
| **Video** | [`python-fer`](https://github.com/justinshenk/fer) | Facial expression recognition using CNN-based FER library. |
| **Audio** | [`Custom-trained CNN model`](https://medium.com/@senirukepasinghe/speech-emotion-recognition-with-cnn-8e3c2cbc8375) | Trained on Mel spectrogram features for emotion classification. |
| **Text** | [`arpanghoshal/EmoRoBERTa`](https://huggingface.co/arpanghoshal/EmoRoBERTa) | Transformer-based text emotion model fine-tuned on GoEmotions dataset. |


## Format and Usage

- File format: **CSV**
- Recommended columns:
  - `Utterance`
  - `Emotion`
  - `File_Name`
  - `Final_Emotion` (JSON: `{ "video": [...], "audio": [...], "text": [...] }`)
  
This dataset is ideal for:
- **Fusion model training**
- **Fine-tuning multimodal emotion models**
- **Benchmarking emotion fusion strategies**
- **Ablation studies on modality importance**


## Citation

References for the original MELD Dataset
- S. Poria, D. Hazarika, N. Majumder, G. Naik, R. Mihalcea, E. Cambria. MELD: A Multimodal Multi-Party Dataset for Emotion Recognition in Conversation (2018).
- Chen, S.Y., Hsu, C.C., Kuo, C.C. and Ku, L.W. EmotionLines: An Emotion Corpus of Multi-Party Conversations. arXiv preprint arXiv:1802.08379 (2018).


## License & Acknowledgments

This dataset is a **derivative work** of MELD, used here for research and educational purposes.  
All credit for the original dataset goes to the **MELD authors** and contributors.