Authors: Adam Bosen¹
¹Boys Town National Research Hospital
Background: Speech recognition in the presence of competing speech is often hindered by a mixture of energetic and informational masking. While performance is typically measured as a single outcome (e.g., percentage correct), this project explores whether analyzing the distribution of recognition accuracy can provide separate estimates of how often each type of masking occurs.
Method: Sentence recognition accuracy in competing speech was analyzed in listeners with normal hearing, with and without target-masker separation cues (spatial and sex mismatch). Data were analyzed using a zero-inflated beta-binomial model. This approach treats recognition as a mixture of two processes: a beta-binomial distribution for when the listener attends to the target, and an “excess zero” rate representing failures to attend the target (informational masking).
Results: When separation cues were absent, the excess zero rate was approximately 17% at the average speech-reception threshold, varying between 5% and 30% across listeners. However, with separation cues present, the excess zero rate was near-zero. Random participant intercepts for the excess zero rate and the beta-binomial mean were uncorrelated, yet both contributed to determining individual speech reception thresholds.
Conclusion: Individuals exhibit varying susceptibility to informational and energetic masking. Zero-inflated beta-binomial models allow these distinct performance aspects to be jointly estimated, providing a more granular understanding of speech recognition challenges in competing speech environments.
