Deep learning reveals what vocal bursts express in different cultures

Area of Science:

• Psychology
• Linguistics
• Artificial Intelligence

Background:

• Human social interactions involve diverse emotional vocalizations, termed 'vocal bursts'.
• Understanding the cross-cultural meanings of these vocal bursts is an emerging field of research.

Purpose of the Study:

• To investigate the shared and culture-specific meanings of vocal bursts.
• To utilize deep learning and large-scale data to analyze emotional vocalizations.

Main Methods:

• Collected data from 4,031 participants across China, India, South Africa, the USA, and Venezuela.
• Participants mimicked and judged the emotional meaning of 2,756 vocal burst recordings.
• Employed a deep neural network to identify acoustic dimensions of vocal expression, disregarding context and speaker identity.

Main Results:

• Discovered 24 distinct acoustic dimensions of vocal expression with emotion-related meanings.
• Found that 79% of the attributed meanings for vocal bursts were preserved across the five countries and three languages.
• The deep neural network successfully predicted culture-specific meanings while ignoring context and speaker.

Conclusions:

• Human emotional vocalizations possess underlying acoustic dimensions with widely shared meanings.
• Cross-cultural emotional vocal communication is significantly more universal than previously understood.
• Deep learning offers a powerful tool for dissecting the complexities of human emotional expression.