maleo-ai/maleo-emotion

Maleo Emotion Audio Dataset Indonesia for Emotion Classification

The Maleo Emotion dataset is a crucial resource developed to address the limited availability of speech emotion datasets in Indonesian. This "in-the-wild" audio dataset is specifically designed for Speech Emotion Recognition (SER) systems, a field with increasing demand across various sectors like customer service, education, and human-computer interaction.

---

Dataset Details

• Dataset Name: Maleo Emotion
• Availability: https://huggingface.co/datasets/maleo-ai/maleo-emotion
• Total Audio Files: 700
• Audio Format: WAV
• Audio Duration: 3 seconds per segment
• Number of Emotion Labels: 7
• Emotion Labels:
  • Marah (Angry)
  • Terkejut (Surprised)
  • Senang (Happy)
  • Takut (Scared)
  • Sedih (Sad)
  • Jijik (Disgusted)
  • Netral (Neutral)
• Files per Emotion Label: 100 audio files for each emotion, ensuring a balanced dataset.
• Data Collection: Audio segments were independently gathered from various video available on the YouTube platform.

---

Motivation and Purpose

The development of the Maleo Emotion Audio Dataset stems from the critical need for Indonesian speech emotion recognition resources. By providing a publicly available and well-annotated dataset, this research aims to:

• Address Resource Scarcity: Fill the gap in Indonesian speech emotion datasets, which is a major challenge for SER system development.
• Facilitate Research: Enable researchers and developers to build, train, and evaluate robust SER systems specifically for the Indonesian language.
• Support Applications: Contribute to the advancement of SER systems for real-world applications in customer service, education, human-computer interaction, and more.

---

Data Processing and Augmentation

The data collection and processing pipeline for Maleo Emotion involved several key stages to ensure its quality and utility:

1. Preprocessing: Raw audio was extracted from YouTube videos and segmented into 3-second WAV files.
2. Feature Extraction: Five main features were extracted from the audio segments, crucial for emotion recognition:
   • Zero Crossing Rate
   • Energy
   • Mel-Frequency Cepstral Coefficients (MFCC)
   • Spectral Roll-off
   • Spectral Flux
3. Augmentation: To improve the generalization power of models trained on this dataset, augmentations were performed, including:
   • Pitch Shifting
   • Noise Injection
   • Time Stretching

---

Baseline Model and Performance

A Convolutional Neural Network (CNN) architecture, implemented using TensorFlow, was used to build a classification model and evaluate the dataset's effectiveness.

• Model Architecture: Convolutional Neural Network (CNN)
• Implementation Framework: TensorFlow
• Evaluation Metric: Accuracy
• Achieved Accuracy: 94.48% on the test data.

The evaluation results demonstrate a balanced performance across all emotion categories, proving the high capability of the developed dataset and model architecture in recognizing emotions from Indonesian speech effectively and relevantly in the local context.

---

Data Structure

The dataset is organized with each emotion label having its own directory containing 100 WAV files. For example:

maleo-emotion/
├── marah/
│   ├── marah_001.wav
│   ├── marah_002.wav
│   └── ...
├── terkejut/
│   ├── terkejut_001.wav
│   ├── terkejut_002.wav
│   └── ...
├── senang/
│   └── ...
├── takut/
│   └── ...
├── sedih/
│   └── ...
├── jijik/
│   └── ...
└── netral/
    └── ...

---

Citation

If you use the Maleo Emotion dataset in your research or project, please cite the following:

BibTeX:

@article{Mardiana2025MaleoEmotion,
  title={{Maleo Emotion Audio Dataset Indonesia untuk Klasifikasi Emosi}},
  author={Mardiana, Ardi and Permana, Sri Mentari Widya Ningrum and Sopiandi, Ii and Bastian, Ade and Irawan, Eka Tresna},
  journal={Unpublished work},
  year={2025},
  url={https://huggingface.co/datasets/maleo-ai/maleo-emotion}
}

APA Style:

Mardiana, A., Permana, S. M. W. N., Sopiandi, I., Bastian, A., & Irawan, E. T. (2025). Maleo Emotion Audio Dataset Indonesia untuk Klasifikasi Emosi. Unpublished work. Retrieved from https://huggingface.co/datasets/maleo-ai/maleo-emotion

---

Contact

For any inquiries or further information regarding this dataset, please contact the authors: Ardi Mardiana, Sri Mentari Widya Ningrum Permana, Ii Sopiandi, Ade Bastian, and Eka Tresna Irawan.