SERENE@DravidianLangTech 2026: Multimodal Approaches for Depression Detection in Dravidian Speech: Acoustic, Spectrogram, and Transformer-Based Models
Year: 2026
Type: conference-paper
Abstract: This paper presents our submission to the Depression Detection in Dravidian Languages shared task at DravidianLangTech 2026.We investigate three complementary approaches for speech-based depression detection in Tamil and Malayalam: (i) acoustic feature engineering using MFCC and prosodic features with a Support Vector Machine (SVM) classifier, (ii) a convolutional neural network (CNN) trained on Mel-spectrogram representations, and (iii) a transformer-based model using Whispergenerated transcripts fine-tuned with XLM-RoBERTa.Experimental results show that acoustic feature-based SVM and spectrogrambased CNN models achieve the strongest performance on both Tamil and Malayalam datasets, while the transformer-based approach also produces competitive results.We further discuss limitations and future research directions.
Institution: Sri Sivasubramaniya Nadar College of Engineering
Language: English
Topic: Emotion and Mood Recognition
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7166883771