Dissociation of human and computer voices in the brain: Evidence for a preattentive gestalt‐like perception

Abstract

Abstract:We investigated the early ("preattentive") cortical processing of voice information, using the so-called "mismatch response." This brain potential allows inferences to be made about the sensory short-term store. Most importantly, the mismatch potential also provides information about the organization of long-term memory traces in the auditory system. Such traces have reliably been reported for phonemes. However, it is unclear whether they also exist for human voice information. To explore this issue, 10 healthy subjects were presented with a single word stimulus uttered by voices of different prototypicality (natural, manipulated, synthetic) in a mismatch experiment (stimulus duration 380 msec, onset-to-onset interval 900 msec). The event-related magnetic fields were recorded by a 148-channel whole-head magnetometer and a source current density modeling of the magnetic field data was performed using a minimum-norm estimate. Each deviating voice signal in a series of standard-voice stimuli evoked a mismatch response that was localized in temporal brain regions bilaterally. Increased mismatch related magnetic flux was observed in response to decreased prototypicality of a presented voice signal, but did not correspond to the acoustic similarity of standard voice and deviant voices. We, therefore, conclude that the mismatch activation predominantly reflects the ecological validity of the voice signals. We further demonstrate that the findings cannot be explained by mere acoustic feature processing, but rather point towards a holistic mapping of the incoming voice signal onto long-term representations in the auditory memory. Hum. Brain Mapping 20:13-21, 2003.

A theoretical framework for CNS arousal

“…This phenomenon, therefore, isolates and identifies a generalized, primitive, non-specific arousal function in the human brain quite separate from all of the high-resolution, evolved sensory pathways that we possess. During systematic laboratory experimentation, evidence for preattentive processing has cropped up in the visual (46,47) and auditory (48)(49)(50) modalities. It declines with age, (51) and its CNS correlates have begun to be studied.…”

Voice perception: Sex, pitch, and the right hemisphere

“…In an earlier study using magnetoencepalography (MEG) [Lattner et al, 2003], we showed that a violation of the listeners' expectations by a non-prototypical voice (e.g., an extraordinarily low female voice) leads to a voice-specific brain response. This response occurs in primary auditory cortices as early as 200 ms after the stimulus onset.…”

Human Communication Investigated With Magnetoencephalography: Speech, Music, and Gestures

“…This way, the pitch-shifted female voice was physically more similar to the standard (diVering in formant structure only) than the natural female voice (diVering in both pitch and formant structure). With this stimulus material, an MEG experiment was performed (Lattner et al, 2003) and the observed mismatch responses were projected onto the brain surface using a minimum norm least squares algorithm (Hämäläinen and Ilmoniemi, 1994;Knösche et al, 1996Knösche et al, , 1999Wang et al, 1992). If the mismatch response depended on short-term memory traces only, it should reflect the physical diVerence of the deviant and the standard, hence the pitch-shifted female voice should evoke the smallest response, the natural female voice should induce a larger response, and the computer-generated female voice should evoke the largest response.…”

“…The bar chart indicates the average current densities per hemisphere in a time window between 100 and 300 ms after stimulus onset. The described diVerences between the various deviants were statistically significant, while there was no hemisphere eVect (adapted from Lattner et al, [2003], Fig. 3, with permissions of John Wiley and Sons Ltd).…”

“…One such method is magnetic field tomography (MFT) (Ioannides et al, 1990). We used the BSCD method to reconstruct the generators of preattentive auditory processing (Lattner et al, 2003) as well as sensory-motor coupling in pianists when they listen to piano music (Haueisen and Knösche, 2001), and the MFT method for localizing brain regions, which are responsible for early syntactic processing (Gross et al, 1998). The BSCD method, as any other distributed source reconstruction method, can yield a quite complex and confusing picture of activity, which results from the overlap of many diVerent neuronal networks, each of which may or may not reflect diVerent functions in a complex cognitive process.…”