← News

ResearchSources: arXiv · 19 Nov 2026

SIGMA model links interview speech, audio and video to estimate depression severity

Researchers report a speaker-aware framework that scores depression severity from semi-clinical interviews, isolating participant speech and predicting questionnaire items jointly for state-of-the-art results.

In plain words

A research team built SIGMA, a system that estimates how severe a person's depression is from recorded clinical-style interviews. It uses only the participant's speech, combines text, audio and video cues, and predicts questionnaire answers together rather than one by one. On a benchmark interview dataset, it reported the best published scores the authors know of.

What was announced

The authors argue that earlier automated approaches have three weaknesses: they keep interviewer speech in the input, which confounds the participant's own cues; they fuse text, audio and video too late, losing fine-grained cross-modal dynamics; and they score symptoms independently, ignoring how symptoms correlate. SIGMA addresses each point. Speaker diarization restricts input to participant speech, cross-attention lets the three modalities condition one another during encoding, and a graded-response head predicts eight questionnaire items jointly through a shared latent severity trait with item-specific ordered thresholds.

On the Extended Distress Analysis Interview Corpus, the team reports state-of-the-art performance, with a concordance correlation coefficient of 0.762, an RMSE of 4.13 and an MAE of 3.30 averaged over five seeds. Ablation tests, where components are removed one at a time, reportedly show that isolating participant speech and predicting symptoms jointly each substantially improve concordance and accuracy. The figures come from the paper's abstract alone, so the underlying study has not been independently reviewed here.

What it does not show

  • Evidence is limited to the paper's abstract; the full study, dataset details and error analyses were not available for review.
  • The reported metrics are the authors' own claims on one benchmark corpus and have not been independently confirmed.
  • The abstract does not describe clinical validation, deployment testing or performance outside the Extended Distress Analysis Interview Corpus.

Sources

Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 20 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.