Soutenance de thèse : Adrielle DE CARVALHO SANTANA

Auditory Evoked Potential (AEP) is a neuroelectric response to an auditory stimulus that reflects the activities of a set of neurons along the pathways of the auditory system. This biopotential is used to aid in the diagnosis of hearing diseases and in the study of auditory processing in the human brain. The improvement of techniques for the acquisition, processing and analysis of biopotentials, makes it possible to deepen our knowledge about the auditory processing. For this, it is interesting to work with more complex stimuli, such as speech, which have more acoustic parameters that vary in time and frequency than the clicks or tones used in traditional audiometric tests. One of the ways to analyze speech processing by the human brain is through the study of categorical perception (CP) of phonemes. The categorical perception of speech sounds consists of the mapping of continuous changes in sound characteristics in discrete perceptual units during phonemic identification. The objective of this work is to investigate the neural correlates of categorical perception of phonemes in Brazilian Portuguese by analyzing the auditory evoked late response (AELR) within the scope of the acoustic characteristics of the phonemes, amplitude and latency of the responses, cortical regions involved, degree of attention to the acoustic task (passive or active) and the physical or psychophysical characteristics of the response. An experiment was carried out with tasks that involved the active and passive categorization of phonemes belonging to two different continua: one based on variations of the voice onset time (VOT), which represents a temporal acoustic cue, and another based on variations of the formant frequencies, a spectral characteristic of the stimulus. AELRs were acquired via electroencephalography (EEG). The analysis of the AELRs was performed in time and time-frequency domains in conjunction with behavioral data obtained from the participants' psychometric curves. In the time-domain the amplitudes and latencies of the AELR components, N1 and P2, were analyzed. In the time-frequency domain, data was represented through discrete wavelet coefficients. To extract the physical and psychophysical representations of the categorical coding we proposed an original regression technique that allows to work with a small amount of observations but large dimensions (a characteristic of our data set) we called, RoLDSIS. The regression coefficients and the N1 and P2 amplitudes and latencies were analyzed through contrast analysis of mixed effects models. The results showed that the categorical perception is affected by the acoustic cue but not by the task and it is coded around the P2 wave latency by the beta-band activity, a brain oscillation related to ambiguity and strength of categorical speech perception. It was observed that participants which present behaviorally stronger CP presented a larger difference between their physical and psychophysical neural representation of the stimuli. We also show that CP occurs when there is not attention to the auditory task but only for the formant based acoustic cue. Hemispheric differences were observed in the speech processing with stronger activity at the left hemisphere. Differences also were observed between frontal and temporal cortical regions where the temporal structures would perform, mainly, a spectrotemporal processing of the stimuli while the frontal structures would perform a psychophysical processing related to the CP. This study compared different dimensions related to categorical speech perception, with stimuli from Brazilian Portuguese, and confirmed many of the results found in the literature for other languages. Acquisitions were made noninvasively using a simple and reproducible protocol allowing that the analyzes we performed can be reproduced for study or evaluation of the categorical perception of participants.

Encadrants :
- Directeur de thèse : Rafael LABOISSIERE - rafael.laboissiereuniv-grenoble-alpes.fr (rafael[dot]laboissiere[at]univ-grenoble-alpes[dot]fr)
- Codirecteur : Hani Camille YEHIA  hanicefala.org (hani[at]cefala[dot]org)

 

Keywords Categorical perception, Discrete Wavelet Transform, Electroencephalography, Acoustic Stimulus, Linear Regression

Read the thesis