Victims and complainants who come to the National Helpline Against Atrocities (NHAA 14566) and its integrated grievance portal are routinely reporting extreme levels of trauma, PTSD, emotional distress and co-existing suicidal thoughts as a result of systemic harassment, caste-based atrocities and other socio-legal emergencies. The intake triage process is under considerable strain due to the sheer volume of calls, the cognitive fatigue of the intake officers, the prevalence of multilingual and codeswitched regional narratives (particularly in Marathi and Hindi), and the inherently subjective nature of manual psychological screening. We present an end-to-end, multimodal, real-time framework for the assessment of stress and trauma which is specifically designed for the NHAA telephony and digital complaint portal channels. The architecture combines an Acoustic Stream and a Textual Natural Language Processing (NLP) Stream through a cross-attentive fusion classifier. The acoustic stream utilises a range of features from the extended Geneva Minimalistic Acoustic Parameter Set (eGeMAPS) to pitch jitter, shimmer, formant dispersion and Mel-Frequency Cepstral Coefficients (MFCCs), which are processed through a Bidirectional Gated Recurrent Unit (BiGRU) to extract the affective biomarkers of voice stress. The textual stream takes the spoken transcripts from the telephony channel and the written complaints from the portal as input and uses a cross-lingual transformer (XLM-RoBERTa / Sentence-BERT) coupled with a zero-shot DeBERTa-v3 Natural Language Inference (NLI) cross-encoder to map the disclosures directly to the DSM-5 criteria for PTSD and the PHQ-9 criteria for depression. The experimental results show that the multimodal fusion approach achieves an overall accuracy of 93.84% (Macro F1: 0.9261, ROC-AUC: 0.9712) in the classification of trauma and distress, which is significantly better than the unimodal text (86.12%) and unimodal voice (81.75%) baselines. The framework also incorporates the Integrated Ethical Approach for Computational Psychiatry (IEACP) and participatory injustice principles with bias-audited risk stratification, explainable TreeSHAP symptom attribution for the caseworkers and zero latency emergency routing in the case of an imminent suicidal crisis.

Sahil Kshirsagar, Dharna Totlani, Vikas Gupta, Gayatri Oman, Shripad S. Bhide, "A Multilingual AI Framework for Real-Time Psychological Stress and Trauma Assessment in Emergency Helpline Systems" International Journal of Scientific Research and Engineering Development, V9(5): Page(2019-2023) September - October 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.