--- title: "SERENE@DravidianLangTech 2026: Multimodal Approaches for" description: "This paper reports the SERENE team's entry in the DravidianLangTech 2026 shared task on detecting depression from Tamil… 11 references." authors: ["TT Pranesh", "K.K.Thamizhmathi", "S Vigneshwaran", "Bharathi B"] year: 2026 venue: "Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages" source: "https://aclanthology.org/2026.dravidianlangtech-1.55/" doi: "10.18653/v1/2026.dravidianlangtech-1.55" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/pranesh-2026-serene-dravidianlangtech-2026-multimodal-approaches/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/pranesh-2026-serene-dravidianlangtech-2026-multimodal-approaches/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/42/) / SERENE@DravidianLangTech 2026: Multimodal Approaches for Depression Detection in Dravidian Speech: Acoustic, Spectrogram, and Transformer-Based Models # SERENE@DravidianLangTech 2026: Multimodal Approaches for Depression Detection in Dravidian Speech: Acoustic, Spectrogram, and Transformer-Based Models **Authors:** [TT Pranesh](https://md.research.elamur.ai/free/en/authors/tt-pranesh/), [K.K.Thamizhmathi](https://md.research.elamur.ai/free/en/authors/k-k-thamizhmathi/), [S Vigneshwaran](https://md.research.elamur.ai/free/en/authors/s-vigneshwaran/) and [Bharathi B](https://md.research.elamur.ai/free/en/authors/bharathi-b/) · **Year:** 2026 · **Venue:** [Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages](https://md.research.elamur.ai/free/en/venues/dravidianlangtech/) ([Workshop on Speech, Vision, and Language Technologies for Dravidian Languages 2026](https://md.research.elamur.ai/free/en/venues/dravidianlangtech/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/2026.dravidianlangtech-1.55/) · [PDF](https://aclanthology.org/2026.dravidianlangtech-1.55.pdf) · [DOI 10.18653/v1/2026.dravidianlangtech-1.55](https://doi.org/10.18653/v1/2026.dravidianlangtech-1.55) · [OpenAlex](https://openalex.org/W7166883771) **Citations:** 0 citations (OpenAlex, 2026-10-14) ## Summary This paper reports the SERENE team's entry in the DravidianLangTech 2026 shared task on detecting depression from Tamil and Malayalam speech, two low-resource languages that have received little attention in this area. Each utterance is classified as Depressed or Non-depressed. The data are small: 1,888 Malayalam and 1,534 Tamil clips from only 8 and 9 speakers respectively, split about 90/10 into training and test sets. The authors compare three pipelines: - an RBF-kernel SVM on 33 handcrafted Librosa features, - a compact CNN on Mel-spectrogram images, - XLM-RoBERTa-base fine-tuned on Whisper medium transcripts. The SVM and CNN score perfectly on both languages, and the text model reaches 0.96 on Tamil. The SVM run took first place in both tracks. The authors warn that these results are probably inflated. The same speakers appear in training and test data, and the two classes differ in recording conditions and sampling rates. The scores therefore say more about the benchmark's weaknesses than about real-world diagnostic value, which makes speaker-independent evaluation a clear priority. ## Key points - **Dataset:** - Malayalam has 888 depressed and 1,000 non-depressed samples (1,688 train / 200 test). - Tamil has 534 depressed and 1,000 non-depressed samples (1,374 train / 160 test). - **Acoustic SVM features and settings:** - The 33-dimensional vector combines the mean and standard deviation of 13 MFCCs with pitch statistics, energy statistics and silence ratio. - The SVM's regularization parameter C is set to 10, its kernel coefficient γ uses the "scale" option, and class weights are adjusted so both labels count equally. - **CNN setup:** - Input is the first 5 seconds of audio, converted to dB-scaled Mel-spectrograms and resized to 128×128. - The network has two convolutional layers (32 and 64 filters) and a sigmoid output, trained with Adam for 20 epochs. - **Text model setup:** XLM-RoBERTa-base is fully fine-tuned on Whisper medium transcripts with learning rate 2e-5, batch size 8 and 6 epochs. - **Results:** - Acc, Macro-P, Macro-R and Macro-F1 are all 1.00 for SVM and CNN on both languages. - Whisper + XLM-R scores 0.96 on Tamil; no Malayalam result is reported for it. - Code is published at github.com/Pranesh4950/Depression-Detection-Task. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* ## Institutions - [Sri Sivasubramaniya Nadar College of Engineering](https://md.research.elamur.ai/free/en/universities/sri-sivasubramaniya-nadar-college-of-engineering/) — University, Chennai, India · 10 works · 9 citations 📚 Shelves (7): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 1,552 works · [💭 Experimental and Cognitive Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/experimental-and-cognitive-psychology/) · 870 works · [💭 Social Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/social-psychology/) · 306 works 🧭 Topics (2): [🖼️ Vision & language](https://md.research.elamur.ai/free/en/topics/multimodal/) · 657 works · [🎙️ Speech & spoken language](https://md.research.elamur.ai/free/en/topics/speech/) · 480 works ## Cite Cite as: TT Pranesh, K.K.Thamizhmathi, S Vigneshwaran and Bharathi B. 2026. SERENE@DravidianLangTech 2026: Multimodal Approaches for Depression Detection in Dravidian Speech: Acoustic, Spectrogram, and Transformer-Based Models. Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{pranesh2026serene, title = {{SERENE@DravidianLangTech 2026: Multimodal Approaches for Depression Detection in Dravidian Speech: Acoustic, Spectrogram, and Transformer-Based Models}}, author = {TT Pranesh, K.K.Thamizhmathi, S Vigneshwaran and Bharathi B}, booktitle = {Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages}, year = {2026}, doi = {10.18653/v1/2026.dravidianlangtech-1.55}, url = {https://doi.org/10.18653/v1/2026.dravidianlangtech-1.55} } ``` - [← previous Sequence Labeling for Constituent Parsing: A Comparative Study and Encoding Innovations](https://md.research.elamur.ai/free/en/papers/roca-2026-sequence-labeling-constituent-parsing-comparative/) - [next → SERM: Self-Evolving Relevance Model with Agent-Driven Learning from Massive Query Streams](https://md.research.elamur.ai/free/en/papers/wang-2026-serm-self-evolving-relevance-model/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/42/) ## References 11 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. A. Anilkumar, Jyothish Lal, B. Premjith, and Bharathi Raja Chakravarthi. 2026. Dravlangguard: A multimodal approach for hate speech detection in dravidian social media. In Speech and Language Technologies for Low-Resource Languages, volume 2656 of Communications in Computer and Information Science,… [find](https://scholar.google.com/scholar?q=A%2E+Anilkumar%2C+Jyothish+Lal%2C+B%2E+Premjith%2C+and+Bharathi+Raja+Chakravarthi%2E+2026%2E+Dravlangguard%3A+A+multimodal+approach+for+hate+speech+detection+in+dravidian+social+media%2E+In+Speech+and+Language+Technolo) 2. Arnab Kumar Das and Ruchira Naskar. 2024. A deep learning model for depression detection based on MFCC and CNN-generated spectrogram features. Biomedical Signal Processing and Control, 90:105898. [find](https://scholar.google.com/scholar?q=Arnab+Kumar+Das+and+Ruchira+Naskar%2E+2024%2E+A+deep+learning+model+for+depression+detection+based+on+MFCC+and+CNN%2Dgenerated+spectrogram+features%2E+Biomedical+Signal+Processing+and+Control%2C+90%3A105898%2E) 3. N. R. Ramalakshmi, Meghna Ganesh Kumar, and S. Raghavi. 2024. Automated depression detection using CNN and transformer-based pre-trained language models. In Proceedings of the International Conference on Computer, Communication, and Signal Processing, pages 42–53. Springer. [find](https://scholar.google.com/scholar?q=N%2E+R%2E+Ramalakshmi%2C+Meghna+Ganesh+Kumar%2C+and+S%2E+Raghavi%2E+2024%2E+Automated+depression+detection+using+CNN+and+transformer%2Dbased+pre%2Dtrained+language+models%2E+In+Proceedings+of+the+International+Conference) 4. Adrián Vázquez-Romero and Ascensión Gallardo-Antolín. 2020. Automatic detection of depression in speech using ensemble convolutional neural networks. Entropy, 22(6):688. [find](https://scholar.google.com/scholar?q=Adri%C3%A1n+V%C3%A1zquez%2DRomero+and+Ascensi%C3%B3n+Gallardo%2DAntol%C3%ADn%2E+2020%2E+Automatic+detection+of+depression+in+speech+using+ensemble+convolutional+neural+networks%2E+Entropy%2C+22%286%29%3A688%2E) 5. Jyothish Lal G, Premjith B, Bharathi Raja Chakravarthi, Saranya Rajiakodi, Durairaj Thenmozhi, and Prasanna Kumar Kumaresan. 2026. Shared task on depression detection from malayalam and tamil speech data. In Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian… [find](https://scholar.google.com/scholar?q=Jyothish+Lal+G%2C+Premjith+B%2C+Bharathi+Raja+Chakravarthi%2C+Saranya+Rajiakodi%2C+Durairaj+Thenmozhi%2C+and+Prasanna+Kumar+Kumaresan%2E+2026%2E+Shared+task+on+depression+detection+from+malayalam+and+tamil+speech+d) 6. Lucía Gómez-Zaragozá, Javier Marín-Morales, Mariano Alcañiz, and Mohammad Soleymani. 2025. Speech and text foundation models for depression detection: Cross-task and cross-language evaluation. In Proceedings of Interspeech 2025, pages 5253–5257. [find](https://scholar.google.com/scholar?q=Luc%C3%ADa+G%C3%B3mez%2DZaragoz%C3%A1%2C+Javier+Mar%C3%ADn%2DMorales%2C+Mariano+Alca%C3%B1iz%2C+and+Mohammad+Soleymani%2E+2025%2E+Speech+and+text+foundation+models+for+depression+detection%3A+Cross%2Dtask+and+cross%2Dlanguage+evaluation%2E+In+Proc) 7. Ah Young Kim, Eun Hye Jang, Seung-Hwan Lee, Kwang-Yeon Choi, Jeon Gue Park, and Hyun-Chool Shin. 2023. Automatic depression detection using smartphone-based text-dependent speech signals: A deep convolutional neural network approach. Journal of Medical Internet Research, 25:e34474. [find](https://scholar.google.com/scholar?q=Ah+Young+Kim%2C+Eun+Hye+Jang%2C+Seung%2DHwan+Lee%2C+Kwang%2DYeon+Choi%2C+Jeon+Gue+Park%2C+and+Hyun%2DChool+Shin%2E+2023%2E+Automatic+depression+detection+using+smartphone%2Dbased+text%2Ddependent+speech+signals%3A+A+deep+convo) 8. A. Kritika, S. Meenakshy, Arya Palackal Shijish, Riya Rajeev, and Jyothish Lal. 2025. Dravimood: Speech-based depression classification in dravidian languages using feature fusion and deep learning. In Proceedings of the Fourth International Conference on Speech and Language Technologies for Low-Res… [find](https://scholar.google.com/scholar?q=A%2E+Kritika%2C+S%2E+Meenakshy%2C+Arya+Palackal+Shijish%2C+Riya+Rajeev%2C+and+Jyothish+Lal%2E+2025%2E+Dravimood%3A+Speech%2Dbased+depression+classification+in+dravidian+languages+using+feature+fusion+and+deep+learning%2E+I) 9. Mariia Nykoniuk, Oleh Basystiuk, Nataliya Shakhovska, and Nataliia Melnykova. 2025. Multimodal data fusion for depression detection approach. Computation, 13(1):9. [find](https://scholar.google.com/scholar?q=Mariia+Nykoniuk%2C+Oleh+Basystiuk%2C+Nataliya+Shakhovska%2C+and+Nataliia+Melnykova%2E+2025%2E+Multimodal+data+fusion+for+depression+detection+approach%2E+Computation%2C+13%281%29%3A9%2E) 10. B. Premjith, G. Jyothish, V. Sowmya, and B. Bharathi. 2024. Findings of the shared task on multimodal social media data analysis in dravidian languages (msmda-dl)@ dravidianlangtech 2024. In Proceedings of the Fourth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages, page… [find](https://scholar.google.com/scholar?q=B%2E+Premjith%2C+G%2E+Jyothish%2C+V%2E+Sowmya%2C+and+B%2E+Bharathi%2E+2024%2E+Findings+of+the+shared+task+on+multimodal+social+media+data+analysis+in+dravidian+languages+%28msmda%2Ddl%29%40+dravidianlangtech+2024%2E+In+Proceedin) 11. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2023. [Robust speech recognition via large-scale weak supervision](https://md.research.elamur.ai/free/en/papers/radford-2022-robust-speech-recognition-large-scale/). In Proceedings of the International Conference on Machine Learning (ICML), pages 28492– 28518. PMLR. *Updated 1 November 2026*