Contribution of Stimulus Variability to Word Recognition in Noise Versus Two-Talker Speech for School-Age Children and Adults

Abstract

Background: Speech-in-speech recognition scores tend to be more variable than the speech-in-noise recognition scores, both within and across listeners. This variability could be due to listener factors, such as individual differences in audibility or susceptibility to informational masking. It could also be due to stimulus variability, with some speech-in-speech samples posing more of a challenge than others. The purpose of this experiment was to test two hypotheses: (1) that stimulus variability affects adults’ word recognition in a two-talker speech masker and (2) that stimulus variability plays a smaller role in children’s performance due to relatively greater contributions of listener factors. Methods: Listeners were children (5 to 10 years) and adults (18 to 41 years) with normal hearing. Target speech was a corpus of 30 disyllabic words, each associated with an unambiguous illustration. Maskers were 30 samples of either two-talker speech or speech-shaped noise. The task was a four-alternative forced choice. Speech reception thresholds were measured adaptively, and those results were used to determine the signal-to-noise ratio associated with ≈65% correct for each listener and masker. Two 30-word blocks of fixed-level testing were then completed in each of the two conditions: (1) with the target-masker pairs randomly assigned prior to each block and (2) with frozen target-masker pairs. Results: Speech reception thresholds were lower for adults than for children, particularly for the two-talker speech masker. Listener responses in fixed-level testing were evaluated for consistency across listeners. Target sample was the best predictor of performance in the speech-shaped noise masker for both the random and frozen conditions. In contrast, both the target and masker samples affected performance in the two-talker masker. Results were qualitatively similar for children and adults, and the pattern of performance across stimulus samples was consistent, with differences in masked target audibility in both age groups. Conclusions: Although word recognition in speech-shaped noise differed consistently across target words, recognition in a two-talker speech masker depended on both the target and masker samples. These stimulus effects are broadly consistent with a simple model of masked target audibility. Although variability in speech-in-speech recognition is often thought to reflect differences in informational masking, the present results suggest that variability in energetic masking across stimuli can play an important role in performance.

Sex-mismatch benefit for speech-in-speech recognition by pediatric and adult cochlear implant users

“…Different types of masker speech have been utilized in previous studies, from single streams of competing speech to multi-talker maskers ( Cullington and Zeng, 2008 ; Meister et al , 2020 ; Stickney et al , 2004 ; Tao et al , 2018 ). Informational masking can differ markedly across conditions and across stimuli within conditions for listeners with normal hearing ( Buss et al , 2021 ; Freyman et al , 2007 ) and for CI users ( Cullington and Zeng, 2008 ). It is possible that discrepancies in prior data on the sex-mismatch benefit in CI users may reflect differences in informational masking in the sex-matched baseline condition ( Meister et al , 2020 ).…”

Band importance for speech-in-speech recognition

“… Buss et al (2020) measured word recognition in a four-alternative forced-choice (4AFC) task. Targets and response alternatives were drawn from a set of 30 disyllabic words.…”

“…Across listeners, TMRs used for fixed-level testing were −25 to –10 dB [mean (M) = –20 dB] for the two-talker masker and −16 to –10 dB (M = –13 dB) for the speech-shaped noise, resulting in mean performance near 65% correct. Buss et al (2020) evaluated the glimpsing opportunities available for these stimuli and argued that listeners required more audible glimpses to recognize speech-in-speech than speech-in-noise. This result indicates that the two-talker speech masker was associated with informational masking even though thresholds were lower than those in the speech-shaped noise masker.…”

“…This report describes an approach for estimating BIFs by evaluating the relationship between audibility of speech cues in restricted frequency bands and correctness of listener response across stimulus samples of unmodified, full-bandwidth speech-in-noise and speech-in-speech. A preliminary test of this approach was undertaken using data reported by Buss et al (2020) for adults with normal hearing. That study measured word recognition in speech-shaped noise and in two-talker speech, using either reproducible target-plus-masker stimuli (described as frozen) or randomly paired stimuli.…”

“…The present report uses the data of Buss et al (2020) to evaluate the feasibility of estimating BIFs using unmodified, full-bandwidth stimuli. We predicted that the BIFs for speech-in-noise obtained using these methods would resemble published functions estimated using low-pass and high-pass filtered speech ( Fletcher and Galt, 1950 ; French and Steinberg, 1947 ).…”