Predicting couple therapy outcomes based on speech acoustic features

Area of Science:

• Computational psychology
• Speech processing
• Relationship science

Background:

• Automated prediction of marital outcomes in couples therapy is challenging but valuable for clinical psychologists.
• Computational methods using observable behaviors from spouse conversations offer objective insights into relationship dynamics.

Purpose of the Study:

• To investigate if acoustic characteristics of spoken interactions in clinically distressed couples can predict therapy outcomes.
• To assess the efficacy of vocal intonation and intensity as features for binary (improvement/no improvement) and multiclass (levels of change) relationship status recognition.

Main Methods:

• Utilized acoustic speech signal characteristics (intonation, intensity) from each interlocutor, both independently and relationally.
• Compared prediction performance against features derived from standardized, human-rated behavioral codes of relationship dynamics.
• Employed data from a longitudinal clinical study of distressed couples.

Main Results:

• Acoustic-based predictions of relationship outcomes were comparable or superior to those using human-rated behavioral codes.
• Combining acoustic features with manually coded behavioral features generally improved prediction performance.
• Analyzing vocal properties relationally (interlocutors' interactions) was more effective than analyzing them in isolation.

Conclusions:

• Vocal acoustics offer a powerful, objective tool for assessing and predicting couples therapy outcomes.
• The interplay between interlocutors' vocal characteristics is crucial for accurate automated prediction of relationship dynamics.
• Acoustic and behavioral features are complementary, with their integration yielding superior predictive insights.