What the study found
The study found that naturalistic speech in people with schizophrenia carried specific, interpretable signs of symptom severity. The speech-based models could estimate symptoms with clinically meaningful accuracy and these patterns were replicated in a U.S. cohort.
Why the authors say this matters
The authors conclude that speech-based modeling provides a basis for scalable, low-burden tools for real-time monitoring in psychosis. They also state that quantitative speech analysis is not yet integrated into clinical care, despite speech being used in psychiatric assessment.
What the researchers tested
The researchers conducted a longitudinal, multicenter cohort study in adult patients with schizophrenia spectrum disorders in the Netherlands, with replication in a U.S. cohort. They analyzed repeated speech and clinical assessments, using the Positive and Negative Syndrome Scale (PANSS) in the Dutch cohort and the Brief Psychiatric Rating Scale in the U.S. cohort, and converted speech into AI-derived voice and text features reduced with principal component analysis.
What worked and what didn't
In the Dutch cohort, item-level symptom estimates were within less than 1 point on a 1-to-7 scale, which the authors describe as clinically meaningful. Models were associated with PANSS positive and negative subscale scores, and in the U.S. replication cohort, thought disturbance and withdrawal scores were also associated with similar error levels and explained variance; negative symptoms were linked with reduced speech output and flatter acoustic profiles, while positive symptoms were linked with longer utterances and altered discourse organization.
What to keep in mind
The summary does not describe detailed limitations beyond the study being conducted in specific cohorts in the Netherlands and the U.S. The findings are based on cohort data and repeated assessments, so the abstract does not claim that the models are ready for routine clinical use.
Key points
- Speech-based models detected individual psychotic symptoms with clinically meaningful accuracy in the Dutch cohort.
- The Dutch cohort included 773 speech recordings from 356 participants; the U.S. replication cohort included 165 recordings from 72 participants.
- PANSS positive and negative subscale scores were associated with the speech models in the Dutch cohort.
- In the U.S. cohort, thought disturbance and withdrawal scores showed similar associations.
- Negative symptoms were linked to reduced speech output and flatter acoustic profiles.
- Positive symptoms were linked to longer utterances and altered discourse organization.
Disclosure
- Research title:
- Speech features tracked symptom severity in schizophrenia
- Authors:
- Silvia Ciampelli, Janna N. de Boer, S Koops, Evan Troelstra, Almut Jebens, Jan-Bernard C. Marsman, AJ Smit, Amir Hossein Nikzad, Ryan Partlan, Philipp Homan, Wolfram Hinzen, Sunny X. Tang, Iris E. C. Sommer
- Institutions:
- ETH Zurich, Feinstein Institute for Medical Research, Feinstein Institute for Medical Research, Feinstein Institute for Medical Research, Institució Catalana de Recerca i Estudis Avançats, Karakter, Northwell Health, Northwell Health, Northwell Health, Universitat Pompeu Fabra, University Medical Center Groningen, University Medical Center Groningen, University Medical Center Groningen, University Medical Center Groningen, University Medical Center Groningen, University Medical Center Groningen, University Medical Center Groningen, University Medical Center Groningen, University of Zurich
- Publication date:
- 2026-06-25
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with scholarly research without reading academic papers — one filtered digest, every Friday.