Abstract
Humans are remarkably skilled at listening to one speaker out of an acoustic mixture of multiple speech sources, even in the absence of binaural cues. Previous research on the neural representations underlying this ability suggests that the auditory cortex primarily represents only the unsegregated acoustic mixture in its early responses, and then selectively processes features of the attended speech at longer latencies (from ~85 ms). The mechanism by which the attended source signal is segregated from the mixture, however, and to what degree an ignored source may also be segregated and separately processed, is not understood. We show here, in human magnetoencephalographic responses to a two-talker mixture, an early neural representation of acoustic onsets in the ignored speech source, over and above onsets of the mixture and the attended source. This suggests that the auditory cortex initially reconstructs acoustic onsets belonging to any speech source, critically, even when those onsets are acoustically masked by another source. Overt onsets in the unseparated acoustic mixture were processed with a lower latency (~70 ms) than masked onsets in either source (~90 ms), suggesting a neural processing cost to the recovery of the masked onsets. Because acoustic onsets precede sustained source-specific information in the acoustic spectrogram, these representations of onsets are cues available for subsequent processing, including full stream segregation. Furthermore, these findings suggest that even bottom-up saliency of objects in the auditory background may rely on active cortical processing, explaining several behavioral effects of background speech.
Significance Statement The ability to comprehend speech in the presence of multiple talkers is required frequently in daily life, and yet it is compromised in a variety of populations, for example in healthy aging. Here we address a longstanding question concerning the neural mechanisms supporting this ability: to what extent does the auditory cortex process and represent an interfering speech signal despite the fact that it is not being attended? We find that auditory cortex not only represents acoustic onsets in an ignored speech source, it does so even when those onsets are masked by the attended talker. This suggests that auditory cortex reconstructs and processes acoustic features of ignored speech, even in its effort to selectively process the attended speech.
Author contributions J.Z.S. and L.E.H. designed experiment and secured funding. C.B. and J.Z.S. analyzed data and wrote the manuscript. A.J., C.B. and J.Z.S. performed simulations.
Footnotes
↵* christianbrodbeck{at}me.com