Articulatory movements modulate auditory responses to speech

Abstract

Production of actions is highly dependent on concurrent sensory information. In speech production, for example, movement of the articulators is guided by both auditory and somatosensory input. It has been demonstrated in non-human primates that self-produced vocalizations and those of others are differentially processed in the temporal cortex. The aim of the current study was to investigate how auditory and motor responses differ for self-produced and externally produced speech. Using functional neuroimaging, subjects were asked to produce sentences aloud, to silently mouth while listening to a different speaker producing the same sentence, to passively listen to sentences being read aloud, or to read sentences silently. We show that that separate regions of the superior temporal cortex display distinct response profiles to speaking aloud, mouthing while listening, and passive listening. Responses in anterior superior temporal cortices in both hemispheres are greater for passive listening compared with both mouthing while listening, and speaking aloud. This is the first demonstration that articulation, whether or not it has auditory consequences, modulates responses of the dorsolateral temporal cortex. In contrast posterior regions of the superior temporal cortex are recruited during both articulation conditions. In dorsal regions of the posterior superior temporal gyrus, responses to mouthing and reading aloud were equivalent, and in more ventral posterior superior temporal sulcus, responses were greater for reading aloud compared with mouthing while listening. These data demonstrate an anterior-posterior division of superior temporal regions where anterior fields are suppressed during motor output, potentially for the purpose of enhanced detection of the speech of others. We suggest posterior fields are engaged in auditory processing for the guidance of articulation by auditory information.

Fig. 1

Speech production, silent articulation and passively listening compared with silent reading. When compared with covert reading (ReadSilently_(novoice)), speech production [ReadAloud_(ownvoice)] is associated with activity in bilateral middle and posterior superior temporal gyri with more distributed activity in the right, and large clusters comprising peaks in ventral somatosensory and premotor and primary motor cortices (1a). Silent articulation with passive listening [MouthSilently_(othervoice)] is a condition where the motor output and auditory input is comparable to normal speech production but the auditory input and motor output are incongruent, i.e. the auditory input is not a direct result of the motor output. Compared with covert reading, this condition was associated with a very similar pattern of activity to normal speech production, including middle STG and ventral somatosensory and motor areas. Visual inspection indicates that activity in the right hemisphere extended posteriorly compared with the [ReadAloud_(ownvoice)] condition seen in the top panel. Finally passive listening [ReadSilently_(othervoice)] was associated with significant BOLD activity in dorsolateral temporal cortex in both hemispheres but again, with more distributed activity in the left. These contrasts are all corrected using a family wise error correction at a threshold of p < 0.05, with a 20 voxel cluster threshold. For full lists of significant peaks see Table 1.

Fig. 2

Passive listening was associated with increased activity in STG and IPL compared with both motor output conditions despite comparable auditory input. All three main conditions, ReadAloud_(ownvoice), MouthSilently_(othervoice) and ReadSilently_(othervoice) involved comparable auditory input. In order to look at how motor production modulates sensory processing, BOLD activity during passive listening was compared with both the auditory motor conditions: ReadSilently_(othervoice) compared with ReadAloud_(ownvoice) was associated with significant activity in middle superior temporal gyri and inferior parietal cortices in both hemispheres (Fig. 2a, red outline). ReadSilently_(othervoice) compared with MouthSilently_(othervoice) was associated with activity in the same regions but extended to inferior frontal gyrus and with a peak in left medial STG/ parietal operculum (Fig. 2a, yellow outline). The reverse contrasts revealed widespread activity in significant activity in bilateral ventral motor, premotor and somatosensory cortices, inferior parietal cortex, inferior frontal cortex and supplementary motor area (Fig. 2b, red and yellow outlines). In order to look directly at the overlap between activity greater for listening compared with the two auditory motor conditions seen in Fig. 3a, a null conjunction was performed of [ReadSilently_(othervoice) > ReadAloud_(ownvoice)] and [ReadSilently_(othervoice) > MouthSilently_(othervoice)] using a masking threshold p < 0.001. This revealed significant activity common to both comparisons in middle to posterior STG and inferior parietal cortices in both hemispheres (Fig. 2c). The parameter estimates for these four clusters were extracted and are plotted in the bottom three panels (Figs. 3d–g) for the contrasts [MouthSilently_(othervoice) > ReadSilently_(novoice)], [ReadAloud_(ownvoice) > ReadSilently_(novoice)], [ReadSilently_(othervoice) > ReadSilently_(novoice)], respectively. These plots demonstrate that inferior parietal regions that respond preferentially to passive listening are suppressed for normal speech (Graphs d and g), but regions in the superior temporal cortex that preferentially respond during passive listening are also active during the two production conditions, but to a lesser degree (Graphs e and f). Despite comparable auditory input, there is more activity in these four regions during passive listening than articulation. This indicates that something about producing a motor articulatory output, whether it be silent movement or not, is modulating activity in these regions (all maps are thresholded at p < 0.005, cluster threshold 20).

Fig. 3

Comparison of auditory motor conditions differentiates between aspects of the auditory pathway that encode self-generated vocalizations. The congruent and incongruent speech conditions are formed of normal speech production [ReadAloud_(ownvoice)], or silent articulation while listening to the same sentence, spoken by someone else played back at the same time [MouthSilently_(othervoice)], respectively. A direct comparison of the two [ReadAloud_(ownvoice) > MouthSilently_(othervoice] revealed significant activations of widespread motor cortices (premotor, inferior frontal, supplementary motor and anterior insula), superior temporal cortex and occipital cortex, (Fig. 3a). The opposite contrast, [MouthSilently_(othervoice > ReadAloud_(ownvoice)], revealed significant activity in bilateral inferior parietal cortex, including both supramarginal and angular gyri (3b). Mean parameter estimates were extracted for the inferior parietal and temporal clusters and are shown in the bottom two panels (Fig. 3c), demonstrating that in both peaks, activity was not only much less for the [ReadAloud_(ownvoice) condition] than the listening [ReadSilently_(othervoice)] condition but was also below baseline, indicating suppression of activity (all maps are thresholded at p < 0.005, cluster threshold 20).

Fig. 4

Separate clusters in superior temporal cortex respond during speech production and listening. All three main conditions were associated with widespread activity across the dorsolateral temporal cortices in both hemispheres. In order to look at how response profiles differed across conditions in the temporal cortex, Fig. 4 displays only peaks lying in the superior temporal cortex for all conditions, revealing an anterior–posterior distribution. Passive listening was associated with significant activity in the most anterior cluster. A null conjunction of [ReadSilently_(othervoice) > [ReadAloud_(ownvoice)] and [ReadSilently_(othervoice) > [MouthSilently_(othervoice)] is shown in yellow, (graphs 4a and 4b). A render of this cluster on an axial slice demonstrates that this anterior cluster extends from the lateral surface to the medial extent of the superior temporal gyrus. Lying within the more middle temporal aspects of the anterior cluster, on the left medial surface is a small peak (green) which is more active for [MouthSilently_(othervoice)] compared with [ReadAloud_(ownvoice)]; the plot of activity within this region (graph 4c) demonstrates that this is due to activity in this small cluster being less suppressed for the mouthing condition than for normal speech production. A more posterior and inferior region (pSTS, shown in red, graphs d and e) is more active for normal speech production compared with mouthing while listening [ReadAloud_(ownvoice)] > [MouthSilently_(othervoice)]. This pSTS cluster extends medially as can be seen on the axial slice. Finally at the posterior and superior extent of the superior temporal gyrus lies a cluster that is commonly active for both auditory motor conditions, shown in blue (null conjunction of [ReadAloud_(ownvoice)] + [MouthSilently_(othervoice)], p < 0.001, graphs 4f and g). Solid lines indicate the cluster and spheres indicate the peak of each cluster. Spherical regions of interest of 3 mm radius were extracted for each peak coordinate and mean parameter estimates were extracted. These are plotted for each peak in graphs a–g where the y axis represents the mean parameter estimate and the three bars represent the mean parameter estimates for 1: [MouthSilently_{(othervoice) >} ReadSilently_(novoice)] 2: [ReadAloud_(ownvoice) > ReadSilently_(novoice)] and 3: [ReadSilently_(othervoice) > ReadSilently_(novoice)]. All clusters are thresholded at p < 0.005, using a cluster extent of 20 voxels.