Speech unfolds at different time scales. Therefore, neuronal mechanisms involved in speech processing should likewise operate at different (corresponding) time scales. The present study aimed to identify speech units relevant for selecting speech streams in a multi-talker situation. Functional connectivity was extracted from the continuous EEG while young adults detected targets within one stream in the presence of a different, task-irrelevant stream. In two separate groups, either the attended or the ignored stream was manipulated so that it contained intact, word-wise scrambled, syllable-wise scrambled, or spectrally scrambled speech. We found functional brain networks that were sensitive to the difference between the situations when speech was meaningful at sentence vs. at word level, but not between when speech was meaningful at word vs. only valid at syllable level, irrespective of whether the speech units were manipulated in the attended or the ignored stream. These functional brain networks operated in the delta and theta bands corresponding to integrating information from longer time windows. Further, the networks, which could be linked with suppressing information from the to-be-ignored stream included brain areas associated with high-level processing of speech. These results are compatible with late filtering models of auditory attention, as they suggest that the length of intact speech units in the to-be-ignored stream affects processes of attentional selection. However, we found no evidence for speech-to-brain coupling differences as a function of the intact unit of speech in either stream. Thus, although the current results do not rule out that early processes of speech processing affect stream selection in a cocktail party situation, neither do they provide supporting for it.

语音以不同的时间尺度展开。因此，涉及语音处理的神经元机制应同样在不同（相应）的时间尺度上运行。本研究旨在确定与在多通话者情况下选择语音流有关的语音单元。从连续的脑电图中提取功能连接性，而年轻人在存在与任务无关的不同流的情况下在一个流中检测到目标。在两个单独的组中，对话务流或忽略流进行了处理，以使其包含完整的，逐字编码的，音节编码的或频谱编码的语音。我们发现，功能性的大脑网络对以下两种情况之间的差异敏感：当句子在单词层面上有意义时，在单词层面上有意义；而当单词在词语层面上有意义时，则对两者之间的差异不敏感。仅在音节级别有效，与语音单元是在有人值班的还是被忽略的流中操作无关。这些功能性大脑网络在对应于来自较长时间窗口的信息积分的δ和θ波段中运行。此外，可以与抑制来自要忽略的流的信息相关联的网络包括与语音的高级处理相关的大脑区域。这些结果与听觉注意力的后期过滤模型兼容，因为它们表明，将被忽略的流中完整语音单元的长度会影响注意力选择的过程。但是，我们没有发现任何证据表明这两种流中的语音与大脑的耦合差异是完整语音单元的函数。从而，