MANAS-AI Research Programme

Research Programme

Multimodal Affective Neuroscience and Speech Analysis System

In partnership with IHBAS Delhi · King's College London · VIMS Uttar Pradesh

Status: Pre-IEC feasibility pilot in preparation · Google.org AI for Science application submitted May 2026

Why MANAS-AI

The Scientific Gap We Are Addressing

No validated AI tool exists for detecting depression in Hindi-speaking populations using speech. Existing speech emotion recognition models achieve only 58–62% accuracy on Hindi speakers because all corpora use simulated, non-clinical speech. No naturalistic, clinically-labelled Hindi depression speech dataset exists anywhere in the world.

MANAS-AI will create that dataset and develop the first validated AI screening model for Hindi-language depression detection — deployable by community health workers on a smartphone, without requiring a specialist at the point of contact.

Methodology

How MANAS-AI Works

A three-step research-to-deployment pathway

1. Clinical Validation Dataset

250–300 psychiatric OPD patients at IHBAS Delhi and VIMS Uttar Pradesh complete three structured speech tasks (~4 minutes). Each recording is labelled with PHQ-9 score and ICD-10 diagnosis as ground truth.

2. AI Model Development

Acoustic features extracted using openSMILE eGeMAPS and Praat. SVM, Random Forest, and XGBoost classifiers trained on labelled data. Validated against PHQ-9 benchmark (target: ≥75% sensitivity, ≥80% specificity).

3. Community Deployment

Validated model integrated into Medijunction Swasthya Didi platform. Community health workers in Alwar District, Rajasthan conduct 4-minute speech screening using a smartphone. Screen-positive individuals referred for teleconsultation.

Collaborators

Research Partners

MANAS-AI is delivered through a collaboration spanning clinical psychiatry, speech AI methodology, and community field deployment

Institute of Human Behaviour & Allied Sciences (IHBAS)

Delhi — Government of NCT of Delhi

Primary research partner. IEC ethics oversight. Clinical data collection site (~150–200 OPD participants). PI: Prof. Suman Kushwaha.

ihbas.delhi.gov.in

King's College London IoPPN

London, United Kingdom

Scientific collaborator. Dr Nicholas Cummins, Senior Lecturer in Speech Analysis and Responsible AI in Health, advises on acoustic feature methodology for Hindi depression speech detection.

kcl.ac.uk/iopphn

Venkateshwara Institute of Medical Science (VIMS)

Amroha, Uttar Pradesh

Phase 2 clinical validation site (~100 OPD participants). Dr Sandeep Choudhary, MD, Head of Psychiatry, as clinical co-investigator.

vimshospital.edu.in

SAPNA (Society for All-round Progressive Necessities and Action)

Alwar District, Rajasthan

Community field partner. 21 years of community presence in Alwar District. Field deployment infrastructure for Swasthya Didi community health worker programme.

sapnaindia.org

Commitment

Open Science

All outputs from the MANAS-AI programme will be published as open resources:

  • Dataset: CC-BY 4.0 on Zenodo — the first clinical-grade Hindi depression speech corpus globally
  • Model: Apache 2.0 on GitHub
  • Publications: Open Access
  • Methodology: CC-BY replication guide for other research groups
Get in Touch

Research Collaboration

For research collaboration enquiries regarding MANAS-AI, contact:

info@medijunction.co.in

Suggested subject line: MANAS-AI Research Collaboration