18–20 Nov 2026
Bodoland University, Kokrajhar, Assam
Asia/Kolkata timezone

Acoustic Signal Characteristics And Machine Learning-Based Classification Of Anger Intention In Naturalistic Bodo Speech

Not scheduled
20m
Bodoland University, Kokrajhar, Assam

Bodoland University, Kokrajhar, Assam

Debargaon, P.O. - Rangalikhata, J.D. Road, Kokrajhar(BTR), Assam, (India) Pin - 783370
Poster Presentation Track III: Computational Physics, Machine Learning & AI Applications in Physics

Speaker

David Brahma (Department of Computer Science and Technology, Bodoland University, Kokrajhar, 783370, BTC, Assam, India.)

Description

Speech carries measurable acoustic characteristics produced through vocal-fold vibration and resonance in the vocal tract. This study proposes a machine learning approach to classify anger intention in naturalistic Bodo debate speech. Approximately 30 minutes of Bodo speech collected from three debate competition recordings available on YouTube will be used to develop the dataset. The recordings have different durations and contain practical recording variations, including microphone noise, echo, reverberation, occasional distortion, and audience/background sounds. The speech data will be grouped into three classes: High Angry, Moderate Angry, and Neutral/Baseline. Twenty-one acoustic and spectral features will be extracted, including Fundamental Frequency (F0), Root-Mean-Square (RMS) Energy, the first three Formants (F1–F3), Spectral Centroid, Spectral Bandwidth, Zero-Crossing Rate, and 13 Mel-frequency Cepstral Coefficients (MFCCs). These features capture aspects of vocal vibration, acoustic energy, vocal-tract resonance, frequency distribution, and the spectral structure of speech. Three supervised learning algorithms-Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbours (KNN) will be used to classify the three anger-intention categories. Their performance will be assessed using accuracy, precision, recall, F1-score, and confusion matrices. The study aims to examine whether measurable acoustic characteristics, combined with machine learning, can distinguish different levels of expressed anger in naturalistic Bodo speech and contribute to computational research on this low-resourced language.

Keywords: Bodo Speech; Acoustic Signal Characteristics; Anger Intention; Speech Analysis; RMS Energy; MFCC; Machine Learning; SVM; Random Forest; KNN.

Author

David Brahma (Department of Computer Science and Technology, Bodoland University, Kokrajhar, 783370, BTC, Assam, India.)

Co-author

Dr Hem Chandra Das (Department of Computer Science and Technology, Bodoland University, Kokrajhar, 783370, BTC, Assam, India.)

Presentation materials

There are no materials yet.