Mental Health

Automatic speech analysis can predict loneliness

TL;DR

Automatic speech analysis provides a proof-of-concept that loneliness is reflected in altered speech, with a model comprising all speech features from a picture description task significantly predicting loneliness, though no single speech feature emerged as a strong predictor.

Key Findings

A machine learning model comprising all speech features from a picture description task significantly predicted loneliness.

  • Sample consisted of 96 healthy participants (mean age 31.08 years, 53 women)
  • Paralinguistic markers related to prosodic, formant, source, and temporal qualities of speech were extracted
  • The predictive model was validated using machine learning analyses conducted for women and men separately
  • The model did not significantly predict loneliness from the storytelling task, suggesting task type matters

No single speech feature emerged as a strong predictor of loneliness.

  • Multiple paralinguistic marker categories were examined: prosodic, formant, source, and temporal qualities
  • Speech features were also correlated with social anxiety and depression in addition to loneliness
  • The finding implies that loneliness is reflected in a combination of subtle speech characteristics rather than any one dominant feature
  • Authors describe the effects as 'small yet significant'

A combined model including both speech features and psychiatric symptoms provided better predictions of loneliness than psychiatric symptoms alone only in women.

  • Machine learning analyses were conducted separately for women and men
  • The additive value of speech features over psychiatric symptoms was sex-specific, benefiting prediction only in women
  • Psychiatric symptoms included measures of social anxiety and depression alongside loneliness
  • This sex difference suggests differential relationships between speech behavior and loneliness across genders

The predictive power of speech features for loneliness was task-dependent, with the picture description task yielding significant predictions but not the storytelling task.

  • Two speech elicitation tasks were used: a picture description task and a storytelling task
  • Only the picture description task model significantly predicted loneliness
  • The storytelling task model did not predict loneliness, suggesting that structured versus unstructured speech contexts differ in their sensitivity to loneliness-related speech markers
  • Both tasks were administered to the same 96 participants

The authors propose that loneliness-related speech alterations may disrupt social interactions and foster chronicity of loneliness.

  • Loneliness has been demonstrated to exert a detrimental effect on mental and physical health
  • The authors suggest loneliness may impede the formation of new social relationships by altering interactional behavior
  • Speech is described as offering 'a new perspective on how loneliness becomes perceptible to others'
  • The findings are framed as a proof-of-concept for making loneliness 'audible'

What This Means

This research suggests that loneliness leaves detectable traces in how people speak. Scientists recorded 96 healthy adults performing two tasks — describing a picture and telling a story — and used automated software to measure subtle characteristics of their speech, such as rhythm, pitch, voice quality, and timing. They then used machine learning to see whether these speech patterns could predict how lonely participants reported feeling. When analyzing the picture description task, a model combining all speech features together could significantly predict loneliness scores, even though no single speech feature on its own was a particularly strong predictor. Interestingly, the same approach did not work for the storytelling task, suggesting that the type of speaking situation matters. The study also found that combining speech features with measures of psychiatric symptoms like depression and social anxiety improved loneliness predictions beyond symptoms alone — but only in women, not men. This sex difference raises questions about whether loneliness is expressed differently in speech depending on gender. Overall, the effects detected were described as small but statistically meaningful, and the authors frame the work as a proof-of-concept rather than a ready-to-use diagnostic tool. This research matters because loneliness is a growing public health concern linked to serious mental and physical health consequences. This research suggests that automated speech analysis could one day offer a non-invasive, objective way to detect loneliness, potentially helping identify people at risk before loneliness becomes chronic. The findings also hint at a possible mechanism by which loneliness perpetuates itself: if lonely people unconsciously speak in ways that are subtly off-putting or socially disengaging, this could make it harder for them to form new relationships, creating a cycle that is difficult to break.

Have a question about this study?

Citation

Diana Immel, E. Mallick, N. Linz, Simon Barton, René Hurlemann, D. Scheele. (2026). Automatic speech analysis can predict loneliness. Scientific Reports. https://doi.org/10.1038/s41598-026-45965-5