Prediction and statistical evaluation of speech recognition test scores

Area of Science:

• Audiology
• Speech-Language Pathology
• Biostatistics

Background:

• Traditional live-voice speech testing is time-consuming.
• Recorded speech materials offer consistency but may lack flexibility.
• Accurate individual speech test result interpretation in clinical settings requires statistical validation.

Purpose of the Study:

• To develop and evaluate a speech test system combining live-voice flexibility with recorded material consistency.
• To establish a statistical basis for interpreting individual monosyllabic word recognition test results.
• To compare individual speech test performance with average performance for similar hearing loss profiles.

Main Methods:

• A speech test evaluation and presentation system was developed.
• Data from 2609 ears (50-word recognition) and 932 ears (25-word recognition) were collected from three audiology clinics.
• Subjects aged 8-92 years with diverse nonconductive hearing losses were included.

Main Results:

• Absolute speech recognition score distributions in hearing-impaired individuals are highly skewed.
• Transformed scores (rationalized arcsine units) showed differences between individual and predicted scores were normally distributed.
• The standard deviation of these differences was approximately 13 rationalized arcsine units across clinics.

Conclusions:

• The developed system provides efficient speech testing using recorded materials.
• Transformed speech recognition scores allow for statistically sound comparisons with audiogram predictions.
• This facilitates more reliable interpretation of individual hearing test results in audiology clinics.