Continued search for better prediction of aided speech understanding in multi-talker environments

Abstract

To better understand issues of hearing-aid benefit during natural listening, this study examined the added demand placed by the goal of understanding speech over the more typically studied goal of simply recognizing speech sounds. The study compared hearing-aid benefit in two conditions, and examined factors that might account for the observed benefits. In the phonetic condition, listeners needed only identify the correct sound to make a correct response. In the semantic condition, listeners had to understand what they had heard to respond correctly, because the answer did not include any keywords from the spoken speech. Hearing aids provided significant benefit for listeners in the phonetic condition. In the semantic condition on the other hand, there were large inter-individual differences, with many listeners not experiencing any benefit of aiding. Neither a set of cognitive and linguistic tests, nor age, could explain this variability. Furthermore, analysis of psychometric functions showed that enhancement of the target speech fidelity through improvement of signal-to-noise ratio had a larger impact on listeners' performance in the phonetic condition than in the semantic condition. These results demonstrate the importance of incorporating naturalistic elements in the simulation of multi-talker listening for assessing the benefits of intervention in communication success.

Neural Activity during Story Listening Is Synchronized across Individuals Despite Acoustic Masking

“…This is consistent with previous studies investigating how speech intelligibility changes with increasing speech degradation or with masking by short speech utterances, such as disconnected sentences (Akeroyd, 2008;Davis & Johnsrude, 2003;Duncan & Aarts, 2006;Obleser et al, 2007;Wild, Yusuf, et al, 2012). This is also consistent with other studies using indirect measures of speech intelligibility during naturalistic listening, such as using comprehension or fill-in-the-blank questions (Best et al, 2016;Humes & Dubno, 2010;MacPherson & Akeroyd, 2013;Power et al, 2012;Xia et al, 2017). One critical difference is that our paradigm directly measures intelligibility by asking participants to report back phrases just after being heard, which allows for a more direct comparison with previous psychoacoustic work using isolated speech utterances.…”

Serial Recall Predicts Vocoded Sentence Recognition Across Spectral Resolutions

“…Comparing predictors of recognition across speech in noise and vocoded speech would likely distinguish stimulus-specific and general relationships between cognition and speech recognition. Along those lines, speech comprehension tasks, such as those described by Best et al (2016), Daneman and Merikle (1996), and Xia et al (2017), likely rely on a different set of cognitive factors than speech recognition tasks and might be more useful for assessing skills related to quality of life. Finally, although serial recall was a strong predictor of sentence recognition, it is not a pure measure of one latent construct of memory, but rather reflects a mixture of abilities (Unsworth & Engle, 2006).…”

Effects of Age on Cortical Tracking of Word-Level Features of Continuous Competing Speech

“…Moreover, even if the listener can correctly piece together the meaning of the utterance, their subjective confidence may be diminished, potentially "blurring" the predictive processes thought to facilitate perception of upcoming speech (Pickering and Gambi, 2018). As such, behavioral measures that more accurately reflect subjective SIN perception difficulties may require utilization of more realistic, narrative stimuli, and focus on quantifying comprehension, as opposed to simple word or sentence identification (e.g., Xia et al, 2017).…”