Emotions in [a]: a perceptual and acoustic study

Area of Science:

• Speech processing and affective computing
• Human-computer interaction
• Phonetics and acoustics

Background:

• Understanding emotional expression in speech is crucial for human-computer interaction.
• Previous research has explored both human and machine capabilities in emotion recognition from voice.
• The effectiveness of short speech units for emotion discrimination remains an area of interest.

Purpose of the Study:

• To investigate the efficacy of voice quality in conveying distinct emotional content.
• To compare human listeners' and computational methods' ability to discriminate emotions from speech.
• To analyze the impact of speech sample length on emotion recognition accuracy.

Main Methods:

• Nine professional actors (4 female, 5 male) produced speech samples.
• Simulated basic emotions (neutral, sadness, joy, anger, tenderness) within vowel units from Finnish speech.
• Employed automatic discrimination algorithms and human listener evaluations for emotion recognition.

Main Results:

• Automatic emotion discrimination from short speech samples significantly outperformed human recognition.
• Computational methods effectively utilized quantitative acoustic parameters for analyzing vowel-length units.
• Human listeners demonstrated lower accuracy with short vowel units compared to machines.

Conclusions:

• Voice quality can convey discernible emotional content, but human perception is sensitive to sample length.
• Machines demonstrate superior ability in recognizing emotions from brief speech segments.
• Further research may explore optimal speech sample durations for human emotion perception.