Individuals with hearing loss often struggle to correctly recognize speech when other talkers are competing in the background even after being fitted with appropriate amplification. This listening situation introduces two different type of masking, or challenges to hear and/or understand the speech signal of interest. Energetic masking occurs when the presence of the background speech degrades representation of the target in the peripheral auditory system. Informational masking is an inability to listen selectivey to the target in the context of a perceptually similar masker. The objective of this proposal is to determine the acoustic factors that cause variability in informational masking across different target and masker stimuli for speech-on-speech recognition, with the rationale that the knowledge gained will facilitate future advances in assessment and intervention, and eventually support technological advances to improve speech-on-speech recognition for listeners with hearing loss. It has been argued that the most important determinant of informational masking in natural speech, where the content is inherently uncertain, is target/masker similarity. While some of the acoustic features underlying similarity for speech-on-speech recognition are known [e.g., differences in fundamental frequency (F0)], we do not yet understand what all the important features are, or how they interact. Further, acoustic cues that increase target salience are known to reduce informational masking, so similarity is not the only determinant of informational masking.
In Aim 1 the variability in informational masking across different target and one- or two-talker masker speech combinations will be assessed for normal-hearing and hearing-impaired listeners. Acoustic analyses of the target and masker speech will allow us to assess how similarity between the acoustic features of target and masker stimuli impact a listener's ability to segregate competing auditory streams.
Aim 2 will focus on informational masking for stimuli produced with either normal or exaggerated prosody for both normal-hearing and hearing-impaired listeners. Exaggerated prosody is often used to draw the listener's attention to the speech signal. Comparing normal and exaggerated prosody will allow for an expanded range of acoustic variables, so we can more clearly see their effects. In addition, it will test the hypothesis that target salience, not just target/masker similarity, is critical for speech-on-speech masking. The same acoustical analyses used in Aim 1 will be conducted on the stimuli employed in Aim 2.
Both Aims will utilize an ideal-binary mask as a way to quantify informational masking across the different stimulus conditions. Multilevel statistical modeling wil use the acoustic features of the target and masker speech to predict informational masking for different speech-on-speech listening conditions. The possibility that different acoustic features are predictive for different listener groups will be explored and tested in all models.
A multivariate approach will be used to understand the acoustic cues underlying speech perception in competing two-talker speech for both normal-hearing and hearing-impaired listeners. The expectation is that the knowledge gained will facilitate improvements in assessment tools and interventions for listeners with hearing loss. This work is relevant to public health, as competing speech often causes significant difficulty for listeners with hearing loss when trying to understand the speech of one talker, and improvements in speech recognition in this type of listening situation would be of value for the millions of Americans affected by hearing loss.