Shifts in rhythm, pitch, fluency or language can be difficult to notice or quantify, but may contain information about emerging psychiatric problems. Researchers from Erasmus MC and TU Delft are exploring whether artificial intelligence can help make these patterns measurable and whether they could ultimately support earlier recognition of people at increased risk of psychotic disorders. 

Psychotic disorders are among the most disabling mental health conditions and can have a profound impact on people’s lives. Earlier recognition may create opportunities for closer monitoring and timely support. Yet the changes that precede a disorder can be difficult to assess objectively. Within the Convergence Health & Technology Flagship Consultation Room 2030, researchers are therefore investigating automated speech analysis as a potential additional tool for psychiatric assessment.
.

When speech becomes a signal 

Speech disturbances are already considered a promising potential marker of emerging psychiatric disorders, but in clinical practice they are still mainly assessed through observation. 

Using speech recordings from adolescents and young adults, the researchers apply machine-learning techniques to analyze linguistic and acoustic features, including pitch, rhythm and fluency, and explore whether combinations of these characteristics are associated with psychiatric symptoms and risk of psychotic disorders. 

The goal is not to diagnose psychosis from someone’s voice or replace clinical judgement. The researchers are examining whether speech features can complement existing assessments and provide useful information alongside interviews, questionnaires, and clinical observation.
.

Following young people over time 

The work is embedded within the iBerry Study (Investigating Behavioral and Emotional Risk in Rotterdam Youth), a large longitudinal cohort study following adolescents at increased risk of psychopathology. Since 2015, iBerry has followed 1,022 young people from the Rotterdam-Rijnmond region into early adulthood. Every two to three years, participants take part in interviews, questionnaires, neuropsychological tests and biological measurements.  

This longitudinal design allows researchers to investigate how speech characteristics develop alongside mental health over time, rather than looking at speech at only one moment. That may ultimately help determine whether changes in speech can provide useful information before psychiatric problems become more severe.
.

First insights from the speech data 

Several Technical Medicine students have contributed to the research under joint supervision from Erasmus MC and TU Delft. 

One example is a 2025 Technical Medicine MSc thesis using speech from semi-structured iBerry interviews. Machine-learning models analyzed acoustic characteristics such as pitch, rhythm and fluency alongside psychiatric symptom measures, including a psychosis screener. 

The study showed that speech characteristics can be systematically extracted and analyzed from the iBerry data, but acoustic information alone was not sufficient for reliable clinical classification. This points towards the need for larger and more diverse datasets, repeated measurements over time and potentially combining acoustic information with linguistic or other types of data. 

The challenge is therefore not to find one characteristic that can ‘detect psychosis’, but to understand which combinations of subtle signals may provide clinically useful information and how they change over time.  

From research to psychiatric care 

The next steps are to extend the analyses to speech collected during later iBerry follow-up waves, validate the findings in clinical samples from Erasmus MC and publish the first scientific findings. 

The team is also exploring a second application of speech technology: speech-to-text tools for psychological and psychiatric interviews. Rather than identifying risk from speech patterns, these tools could support professionals by converting spoken information into text and helping to make the assessment process more efficient and structured. 
.

Potential impact on mental healthcare 

If validated, speech analysis could provide an objective, non-invasive and scalable addition to existing psychiatric assessments. It could help professionals identify people who may benefit from closer monitoring or earlier support. Speech-to-text technology could contribute in a different way by making psychological and psychiatric interviews more efficient and structured. 

Together, these applications could ultimately contribute to more timely, efficient and accessible mental healthcare.