Speech-based personality prediction using deep learning with acoustic and linguistic embeddings.

Area of Science:

• Psychology
• Computer Science
• Linguistics

Background:

• The Big Five personality traits are a widely accepted model for understanding human personality.
• Assessing personality traditionally relies on self-report questionnaires, which can be subject to biases.
• Computational methods offer novel avenues for objective personality assessment.

Purpose of the Study:

• To develop and validate a novel method for predicting the Big Five personality traits using speech analysis.
• To explore the utility of acoustic and linguistic features in speech for personality prediction.
• To advance the field of computational personality assessment.

Main Methods:

• Collected speech samples and Big Five personality data from 2045 participants.
• Utilized pre-trained convolutional neural networks and transformer models to extract speech embeddings (acoustic and linguistic).
• Employed gradient boosted tree models for personality trait prediction based on combined embeddings.

Main Results:

• Speech analysis effectively predicted Big Five personality traits, with correlation coefficients ranging from 0.26 to 0.39 (disattenuated: 0.39 to 0.60).
• The model demonstrated moderate to high consistency in its predictions, as indicated by intraclass correlations.
• Both the acoustic and linguistic aspects of speech were found to be informative for personality assessment.

Conclusions:

• Personality traits can be reliably predicted from speech samples, highlighting the expressive nature of vocal communication.
• Voice-based personality assessment shows significant potential as a complementary tool in psychological research.
• This study provides new insights into the intricate relationship between speech characteristics and individual personality differences.