Modeling the Dynamics of Attentional Gamma Oscillations During the Encoding Process of Noise-Mixed Speech Signals

원문 초록

The brain's bottom-up loop for processing speech influx involves both the selective attention and the encoding of specific speech information. Previous human studies have found that such attention can be represented by the cortical gamma-rhythm oscillations. However, the underlying mechanisms remain unclear. To address this issue, this paper proposes a neural network model that incorporates speech signal input, the cochlea, the thalamus, and a balanced excitatory-inhibitory cortical neural network, with the aim of connecting real speech signals to brain cortical responses. Using this model, we explored neural oscillation patterns in response to mixed speech stimuli and background noise. The findings revealed that the peak of gamma oscillation decreased as the frequency of the pure-tone stimuli diminished. This suggests a strong correlation and coding role of gamma oscillation peaks in auditory attention. Similar results were confirmed by analyzing the rhythmic oscillations of EEG data in response to pure-tone signals. Further results indicated that dynamic gamma oscillations are involved in the encoding capacity of continuous speech input. The coding entropy of the dynamic series was found to be proportional to the complexity of the content. This suggests that gamma oscillations play multiple roles, not only in sustaining the bottom-up attentional state but also in potentially conveying specific information from external speech inputs. Finally, we found that enhancing the excitatory-inhibitory balance level properly could improve auditory attention. This finding provides a potential endogenous explanation for the dynamic switching process of brain attention in processing auditory signals.