Short-term audibility is a better predictor of vocoded speech-in-speech recognition than long-term target-to-masker ratio

Abstract

Under natural listening conditions, speech is often heard in competing speech. In this case, the audibility of the target speech depends on short-term level fluctuations in both the target and the masker. In this experiment, we tested whether short-term audibility was a better predictor of speech-in-speech recognition than the average long-term target-to-masker ratio. 21 young adults with normal hearing listened to frozen combinations of PRESTO sentences and two-talker masking speech presented at average target-to-masker ratios between +10 and 0 dB in 2 dB steps. Frozen stimuli were vocoded with a 16-channel sine wave vocoder to discretize temporal envelope interactions between target and masker across frequencies. For each frozen stimulus we calculated short-term audibility as the proportion of time the envelope of the target speech was greater than the envelope of the masker within and across vocoder channels. Even though long-term target-to-masker ratios were set to a fixed value for each frozen stimulus, short-term audibility varied substantially across stimuli with the same target-to-masker ratio. Short-term audibility was also better predictor of speech recognition for each combination than long-term target-to-masker ratio. We conclude that short-term audibility quantifies interactions in level between fluctuating targets and maskers, whereas long-term target-to-masker ratio does not.