CENTIVE : Audio Signal Processing Challange

SER in Bahasa Dataset is an open-source resource designed to support research and development in Speech Emotion Recognition (SER), particularly in the Indonesian language context. The dataset contains speech recordings annotated into five types of emotions and was collected under controlled conditions with the participation of recruited actors. All participants were involved with informed consent and ethical clearance procedures to ensure research compliance.

Under this challenge, we provide preprocessed audio features (e.g., MFCC and spectrogram representations) along with raw audio files. With this data, participants are expected to build and optimize deep learning models that can classify emotions from Indonesian speech. The challenge aims to benchmark performance, encourage methodological improvements, and foster collaboration within the SER community.

The dataset used in this study is derived from speech recordings collected during the Evaluation of Lecturer Performance (EDOM) process, where students provide verbal feedback regarding their lecturers. The audio recordings are then annotated according to 5 emotional categories (Neutral, Happy, Surprised, Angry, and Sad) to capture the affective tendencies expressed by students.

To ensure research ethics, the dataset collection was conducted under ethical clearance procedures and every participant signed an informed consent form, allowing their data to be used strictly for research and academic purposes. Any personally identifiable information (PII) has been removed to maintain privacy and confidentiality.

The dataset is organized into three main categories that represent potential roles in emotional dynamics within academic evaluation:

1. Positive emotions (happy, surprised),
2. Negative emotions (angry, sad, disappointed), and
3. Neutral which may reflect objective or less expressive evaluations.

The dataset is openly available and can be accessed through the following GitHub repository (readme.txt):

Speech Emotion Recognition EDOM Dataset

Researchers and practitioners are encouraged to cite the related publications listed in the repository when using this dataset for their studies.

The main mission of this study is to design and train a robust deep learning model for classification tasks in Speech Emotion Recognition (SER) applied to lecturer evaluation (EDOM). Participants are given full flexibility in selecting the methodology, including preprocessing techniques, feature engineering strategies, and model architectures.

The challenge lies in building a single classifier model capable of learning from real counseling or lecture session recordings and accurately identifying emotional states that serve as objective indicators of lecturer performance. Evaluation metrics will include accuracy, precision, recall, and F1-score, which together ensure not only correctness but also reliability and fairness in classification.

To push the limits of current SER approaches, participants are encouraged to explore advanced model architectures (e.g., CNN-LSTM hybrids, attention-based models, or transformer variants), experiment with different audio feature extraction methods (MFCC, LPC, spectral features, etc.), and apply optimization strategies (e.g., feature selection, regularization, hyperparameter tuning) to maximize performance.

Beyond the numerical results, the mission also requires a critical feature analysis and in-depth discussion to explain which emotional patterns are most significant in predicting lecturer evaluation outcomes. The ultimate goal is to demonstrate how a well-designed classifier can serve as an objective complement to traditional subjective evaluations (EDOM) in higher education.

The study is expected to produce the following outcomes:

1. A reliable Speech Emotion Recognition (SER) framework using deep learning tailored for lecturer evaluation (EDOM).
2. High accuracy and balanced performance across precision, recall, and F1-score to ensure fairness in classification.
3. Identification of emotional patterns that correlate with lecturer effectiveness in teaching.
4. An objective complement to traditional subjective lecturer evaluations.
5. Contribution to the advancement of SER applications in higher education quality assurance.

Paper Submission

1. All participants/teams must submit a paper through the official CENTIVE 2025 submission system using the provided template.
2. Please select the Challenge Track and choose Speech Emotion Recognition (SER) for Lecturer Evaluation as the subject.

Presentation Requirement

1. All accepted papers must be presented at CENTIVE 2025.
2. Selected papers will be published in the CENTIVE 2025 list of journals (SINTA 2 / Scopus).

Citation Requirement

As part of this challenge, participants are required to cite the following reference in their paper:

1. Rosita, Y. D., Salsabila, Z., & Pamungkas, A. R. P. (2025). Lecturer Evaluation from the Perspective of Speech Emotion Recognition with Deep Learning, "2025 International Conference on Data Science and Its Applications (ICoDSA), 2025, pp. (on process in publishing), doi: (on process in publishing).
2. Rosita, Y. D., Firmansyah, M. R., & Utami, A. (2025). Exploring bibliometric trends in speech emotion recognition (2020-2024). IAES International Journal of Artificial Intelligence (IJ-AI), 14(4), 3421. https://doi.org/10.11591/ijai.v14.i4.pp3421-3434

Available at: https://github.com/yesydiahrosita/SpeechEmotionRecognitioninBahasa/

Note : Failure to include this citation will result in disqualification from the challenge.

Evaluation

The submitted articles will be evaluated using several criteria:

1. Classification Performance: Measured by Accuracy, Precision, Recall, and F1 Score on the EDOM-based dataset.
2. Emotion Recognition Reliability: Evaluation of how well the model captures speech emotion patterns relevant to lecturer–student interactions.
3. Methodology & Innovation: The novelty of preprocessing, feature extraction, and deep learning methods applied to address the problem.
4. Article Quality: Clarity of writing, depth of analysis, and completeness of results presentation.
5. Comprehensive Insight: Additional points for submissions that provide detailed analysis on educational implications and potential real-world applications.

Award

Two challengers will be awarded for this challenge with benefit as follows:

1. Prize award Rp. 2.000.000
2. Free APC publication in CENTIVE2025 journal partners
3. Certificate and Placard