From acoustic to linguistic analysis of temporal speech structure: acousto-linguistic transformation during speech perception using speech quilts.
Speech perception entails the mapping of the acoustic waveform to linguistic representations. For this mapping to succeed, the speech signal needs to be tracked over various temporal windows at high temporal precision in order to decode linguistic units ranging from phonemes (tens of milliseconds) to sentences (seconds). Here, we tested the hypothesis that cortical processing of speech-specific temporal structure is modulated by higher-level linguistic analysis. Using fMRI, we measured BOLD signal changes to 4-s long speech quilts with variable temporal structure (30, 120, 480, 960 ms segment lengths), as well as natural speech, created from a familiar (English) or foreign (Korean) language. We found evidence for the acoustic analysis of temporal speech properties in superior temporal sulcus (STS): the BOLD signal increased as a function of temporal speech structure in both familiar and foreign languages. However, activity in left inferior gyrus (IFG) revealed evidence for linguistic processing of temporal speech properties: the BOLD signal increased as a function of temporal speech structure only in familiar, but not in foreign speech. Network analyses suggested that left IFG modulates processing of speech-specific temporal structure in primary auditory cortex, which in turn sensitizes processing of speech-specific temporal structure in STS. The results thus reveal a network for acousto-linguistic transformation consisting of primary and non-primary auditory cortex, STS, and left IFG.\n\nSignificance StatementWhere and how the acoustic information contained in complex speech signals is mapped to linguistic information is still not fully explained by current speech/language models. We dissociate acoustic from linguistic analyses of speech by comparing the same acoustic manipulation (varying the extent of temporal speech structure) in two languages (native, foreign). We show that acoustic temporal speech structure is analyzed in superior temporal sulcus (STS), while linguistic information is extracted in left inferior frontal gyrus (IFG). Furthermore, modulation from left IFG enhances sensitivity to temporal speech structure in STS. We propose a model for acousto-linguistic transformation of speech-specific temporal structure in the human brain that can account for these results.