Listeners form average-based representations of individual voice identities

Area of Science:

• Cognitive Psychology
• Auditory Perception
• Speech Processing

Background:

• Voice perception models suggest identity encoding relative to an average or prototype.
• Evidence exists for norm-based coding in voice discrimination, but formation of individual voice identity representations remains unclear.

Purpose of the Study:

• To investigate how abstracted representations of individual voice identities are formed through variable exposure.
• To determine if listeners can infer and utilize average voice representations without direct exposure to them.

Main Methods:

• Two experiments were conducted using three perceptually distinct artificial voice identities with controlled within-person variability.
• Participants learned to recognize these identities presented as ring-shaped distributions lacking central points.
• Testing involved old/new judgments on stimuli within trained (ring) and untrained (center) distributions.

Main Results:

• Listeners demonstrated higher accuracy for stimuli located in the untrained central distribution compared to the trained peripheral (ring) distribution.
• This indicates the formation of an abstracted average representation, inferred from the presented peripheral variations.

Conclusions:

• Abstracted, norm-based representations of individual voice identities can be formed even when the average itself is not directly experienced.
• This challenges previous assumptions and provides new insights into the mechanisms of voice identity encoding and memory.