Acoustic voice variation within and between speakers

Area of Science:

• Acoustic phonetics
• Speech perception
• Speaker recognition

Background:

• Everyday voice quality variability is not well understood.
• Current models of voice perception lack detailed understanding of within-talker acoustic variation.
• Hypotheses based on face perception and cognitive models suggest common and idiosyncratic voice features.

Purpose of the Study:

• To explore within- and between-talker acoustic variation in voice quality.
• To determine how acoustic variation aligns with current voice perception models.
• To investigate the structure of individual and population voice spaces.

Main Methods:

• Principal component analyses (PCA) were used.
• Multiple sentence productions from 100 English speakers (50 female, 50 male) were recorded over three days.
• Twenty-six acoustic variables were measured every 5 ms on vowels and approximants.

Main Results:

• The balance between harmonic amplitudes and inharmonic energy explained the most between-speaker variance (20% females, 22% males).
• Formant frequencies and their variability accounted for an additional 12% of between-speaker variance.
• Most remaining acoustic variance was idiosyncratic, but individual and population voice spaces shared similar acoustic structures.

Conclusions:

• A significant portion of voice variability is unique to each speaker.
• Despite individual differences, the underlying acoustic structure of voice spaces is consistent across speakers.
• Findings have implications for prototype models of voice perception and recognition.