Neural tracking of linguistic information as a measure of speech understanding in noise

Authors: David Meng 1; Angela Wong 1; Jennifer Clemesja 1; Jorge Mejia 1

1 National Acoustic Laboratories

Test tools that can directly tap into central auditory processing are desired to better understand variations in speech-in-noise comprehension among individual Hearing Aid (HA) users and Cochlear Implant (CI) recipients. Human cortical response has been shown to track the linguistic units embedded in connected speech, dissociated from any acoustic cues to them. This neural tracking activity could potentially serve as an index of speech understanding in noise.

In the current study, we introduce a multi-talker babble noise to speech sentences and manipulate its level to test the hypothesis that the tracking response can be directly modulated by the variations in Signal-to-Noise Ratio (SNR) and associated changes in speech intelligibility. Cortical electroencephalography (EEG) responses to running speech at four different SNR levels were measured in nineteen normal hearing participants.

Results showed that brain response to abstract linguistic information in connected speech can be reliably measured using EEG, even when concurrent background noise is present. This neural tracking activity reduced systematically as the level of the multi-talker babble noise went up. Notably, the patterns of reduction were different for responses that correspond to different linguistic units. At syllable level, the sensitivity is mainly driven by changes in acoustic property of speech signal (i.e., SNR). Whereas for larger linguistic units that are dissociated from acoustic cues (i.e., phrases and sentences), the changes in tracking responses are caused by perceived intelligibility which underlines speech understanding.

This dissociation suggests that cerebral responses to linguistic information are directly affected by speech intelligibility, which in turn are powerfully shaped by acoustic cues in speech. These results provide an objective and sensitive neural marker which can potentially lead to clinical utilities for objective assessment of speech understanding in noise.