Researchers from the Krauthammer Lab at the Department of Quantitative Biomedicine have developed an uncertainty-aware multimodal model for speech analysis across the psychosis spectrum. The study integrates acoustic and linguistic signals to improve the prediction of symptom severity in individuals with psychosis and schizotypal traits.
Using German speech data from clinical interviews and narrative tasks, the model dynamically adapts to variability in speech quality and context. By estimating modality-specific uncertainty, it identifies when acoustic or linguistic cues are more reliable, leading to robust and well-calibrated predictions.
The work highlights the value of uncertainty modeling for interpretable and trustworthy speech-based psychiatric assessment, supporting more objective evaluation across clinical and non-clinical populations.
Publication: npj Digital Medicine (2026)
DOI: https://doi.org/10.1038/s41746-025-02309-3