| |
Abstract:
Author:J. N. Kaufman Abstract: Word recognition is possible
for modified speech signals containing coarse frequency resolution,
but lacking spectral detail (Shannon et al., Science, 1995). We
measured cerebral blood oxygenation signals as a function of
spectral information to determine the brain regions involved in
this processing. Ten subjects were studied using high-resolution
fMRI at 3-Tesla. A clustered acquisition technique was used to
avoid effects of acoustic scanner noise. Stimuli were created by
first isolating different spectral regions of spoken pseudowords
through bandpass filtering. Amplitude envelopes were extracted from
these bandpassed signals and used to modulate white noise confined
to the same frequency bands. These signal-modulated noise bands
were combined to create stimuli with 1, 2, 3, or 4 bands of
spectrally limited amplitude information. The resulting stimuli
preserved amplitude characteristics of the original phonemes but
contained varying degrees of spectral resolution. Subjects
responded via key-press to the onset and termination of the
signals. Using regression analysis, foci showing activation
correlated with spectral information content were identified in the
anterolateral aspect of the transverse temporal sulcus bilaterally,
with left hemisphere predominance. This region is immediately
anterolateral to core auditory cortex and may be involved in
resolving spectral peaks characteristic of speech.
|
