The frequencies that convey speech cues differ when listening in noise vs competing speech

What is it about?

Band importance functions characterize how much speech information is contained in different speech frequencies. These functions are used to predict speech recognition under different conditions, such as with hearing aids or hearing protection. When recognizing speech in noise, listeners tend to rely most heavily on cues in the frequency region of 1-3 kHz. This means that audibility of speech bands in the 1-3 kHz region is more important for speech-in-noise recognition than audibility of lower and higher frequencies. Some recent research suggests that band importance functions estimated using background noise may not accurately characterize the cues used when listening in a background of other talkers. Speech-in-speech recognition is often more challenging than speech-in-noise recognition because we rely on acoustic differences between target speech and background sounds to selectively attend to the target. Selectively attending is more difficult when the target and the background have similar features, such as when they are both speech. The purpose of this study was to compare band importance functions across speech-in-noise and speech-in-speech. Listeners were 24 adults (18-41 years) with normal hearing. They were tested on a four-alternative forced-choice word recognition task, and background maskers were either speech-shaped noise or two-talker speech. Band importance functions were estimated using two different methods; the relationship between listeners’ responses and the target audibility in each frequency band was characterized via correlation or logistic regression. These two methods produced similar results. As observed in previous research, the band importance function for speech-in-noise indicated reliance on cues at and above 3 kHz. The novel finding here was that the band importance for speech-in-speech recognition was greatest for low- and mid-frequency cues at and below 2.6 kHz. These results support the idea that band importance differs for speech-in-noise and speech-in-speech and suggests that prediction of speech recognition could be improved by using a band importance function that reflects the listener’s everyday auditory environments.

Featured Image

Why is it important?

This study introduced a new method for estimating band importance for speech-in-speech recognition that uses full-bandwidth stimuli without filtering out frequencies, as previous methods have done. Whereas filtering can modify the cues supporting auditory stream segregation (e.g., across-frequency temporal coherence), full-band speech preserves all the cues available in natural listening conditions. Accurately characterizing band importance functions for speech-in-speech recognition is important because natural listening environments often contain multiple talkers (think of a crowded restaurant or a party), and difficulties understanding speech in a background of other talkers is a common complaint among audiology patients.

Perspectives