Determining the Likelihood Ratio From Perceptual Attributes of Voice

Area of Science:

• Forensic Science
• Acoustics
• Speech Analysis

Background:

• Perceptual evaluation of voice attributes is crucial in forensic speaker comparison.
• Existing methods for likelihood ratio (LR) calculation often rely on acoustic measurements.
• A standardized method for integrating expert perceptual judgments into LR calculation is needed.

Purpose of the Study:

• To develop a method for forensic voice experts (FVEs) to determine the likelihood ratio (LR) from perceptual evaluations of distinctive voice attribute scores.
• To quantify similarity and typicality judgments made by FVEs when comparing voice attributes in pairs.

Main Methods:

• A two-stage scoring method was employed for LR calculation.
• Stage one involved FVEs evaluating 17 perceptual attributes across six vocalic categories using a novel graphical interface for 10 same/different speaker pairs.
• Stage two converted mean scores into LRs using probability density functions derived from listener responses to 1680 speaker pairs.

Main Results:

• The FVEs' responses demonstrated the discriminatory power of attributes and the impact of typicality on the final score.
• Probability density functions were applied to convert scores into LRs, which were then compared against known speaker origins.
• The developed system showed high agreement in evaluating perceptual attributes for voice pair comparisons.

Conclusions:

• The developed system enables a reliable, global, and discriminated evaluation of perceptual voice attributes.
• The method successfully links perceptual evaluations with modern automatic voice analysis techniques.
• FVE responses serve as a benchmark for training and evaluating novice forensic voice experts.