The effect of noise spectrum on speech recognition performance-intensity functions

Articulation theory predicts that a subject's absolute or masked threshold configuration will affect the slope of the speech recognition performance-intensity (P-I) function. This study was carried out to test that prediction. Performance-intensity functions for the Technisonic Studios W-22 recordings were obtained from 12 subjects with normal hearing. Four continuous thermal noise maskers, high-pass (HP) noise, white noise, ANSI noise, and talker-spectrum-matched (TSM) noise, were used to shape threshold. P-I function slopes for the averaged data ranged from about 1.6%/dB in HP noise to about 6.7%/dB in TSM noise. At low to moderate speech intensity levels, the positions and slopes of the P-I functions were accurately estimated by an articulation index-type model that included corrections for subject proficiency and for high- and low-frequency spread of masking. At higher intensity levels, performance was overestimated by the model.