Maximum Expected Information Approach for Improving Efficiency of Categorical Loudness Scaling

Area of Science:

• Audiology
• Psychoacoustics
• Signal Processing

Background:

• Categorical loudness scaling (CLS) provides insights into hearing perception across the dynamic range.
• A multi-category psychometric function (MCPF) model has been established for CLS categories.
• Current CLS measurement procedures face challenges with clinical adoption due to lengthy test times.

Purpose of the Study:

• To extend the MCPF approach using Bayesian inference for adaptive stimulus selection (MCPF-MEI).
• To evaluate the accuracy and reliability of the MCPF-MEI procedure against standardized CLS methods.
• To assess the potential for reducing test time in CLS measurements.

Main Methods:

• Utilized a catalog of potential listener MCPFs with maximum-likelihood estimation.
• Employed Bayesian inference to select stimulus parameters for maximum expected information (MEI).
• Compared the MCPF-MEI adaptive procedure to the standardized CLS procedure (ISO 16832:2006) and a non-adaptive gold standard.

Main Results:

• Reduced test time from approximately 15 minutes to 3 minutes using the MEI-adaptive procedure.
• Achieved test-retest reliability and accuracy comparable to the standardized CLS procedure.
• Computer simulations indicated that MCPF catalog uncertainty limited MEI procedure efficiency.

Conclusions:

• The MCPF-MEI adaptive procedure offers a significantly faster alternative for CLS measurements without compromising accuracy or reliability.
• The MCPF catalog is valuable for maximum-likelihood determination of loudness categories, irrespective of the adaptive tracking method.
• Further optimization of both the MCPF catalog and the adaptive procedure could enhance results.