Perception de la parole

La perception de la parole est le processus par lequel les humains sont capables d'interpréter et de comprendre les sons utilisés dans le langage. L'étude de la perception de la parole est reliée aux champs de la phonétique, de phonologie en linguistique, de psychologie cognitive et de perception en psychologie. Les recherches dans ce domaine essaient de comprendre comment les auditeurs humains reconnaissent les phonèmes (sons de la paroles) ou autres sons tels que la syllabe ou les rimes, et utilisent cette information pour comprendre le langage parlé. Ces recherches ont des applications dans la construction de systèmes informatiques qui puissent reconnaître la parole. Définition Le processus de perception de la parole commence au niveau du signal sonore et des processus d'audition. Après le traitement du signal auditif initial, les phonèmes sont traités pour extraire les indices acoustiques et l'information phonétique. Cette information de la parole peut aussi être utili

Some applications of a priori knowledge in multi-stream HMM and HMM/ANN based ASR

Multi-band ASR was largely inspired by the extremely high level of redundancy in the spectral signal representation which can be inferred from Fletcher's product-of-errors rule for human speech perception. Indeed, the main aim of the multi-band approach is to exploit this redundancy in order to overcome the problem of data mismatch (while making no assumptions about noise type) by focusing recognition on sub-bands estimated to contain reliable, or "clean speech like", data. However, multi-band processing also presents the opportunity to introduce a number of other ideas from phonetics, non-linear phonology and auditory processing into the recognition process. In particular: we can weight sub-bands, or sub-band combinations, according to the most likely frequency range of characteristic features for the phoneme whose presence we are testing for; we can allow some degree of asynchrony between sub-bands, and we can preprocess each sub-band according the kind of acoustic features which we expect to find there. Besides combining sub-band experts, we can also combine multiple full-band experts, where each expert is perhaps suited to extracting complementary sources of speech information, or is robust to different kinds of noise. In this article we present an outline of some of the recent work at IDIAP, and cooperating institutions, in bringing together ideas from different areas of speech science within the framework of multi-stream HMM and HMM/ANN based ASR.

2000

Dissecting self-voice perception: From bone conduction to robotically-induced self-other voice misattribution in healthy listeners

Pavo Orepic

Inspired by the prominent theory suggesting that auditory-verbal hallucinations (AVH) - the sensation of hearing voices without present speakers - arise as misattribution of inner speech towards external agents, my thesis revolved around self-voice perception and the experimental attempts of self-other voice misattribution in a healthy, non-hallucinating population, potentially mimicking the AVH phenomenology. In the first part of my thesis, I investigated behavioral (Study 1) and neural (Study 2) underpinnings of self-other voice discrimination (SOVD). Compared to other self-related processes, self-voice perception has been investigated to a surprisingly lesser extent. Namely, self-voice research has been thwarted by the inability to experimentally match self-voice recordings to the natural sound of our voice that is altered by bone conduction. In a series of five experiments, I showed that this discrepancy can be reduced by presenting self-voice stimuli through a commercial bone conduction headset, thereby rendering self-voice as an essentially multi-modal construct. My data further shows that self-voice recognition differs from the recognition of familiar voices, however, that it still involves some familiarity processing. In addition, I explored the roles other-voice familiarity and acoustic similarity to other voice play in SOVD. Finally, I identified a self-voice specific EEG pattern, around 345 milliseconds after stimulus onset, that followed the initial auditory cortex activation, discriminated the self-voice from the voice of another unfamiliar person, and activated an extended network involving the cingulate cortex, insula, and medial temporal lobe structures. Moreover, this network was recruited less frequently with self-voices presented through bone conduction, and the occurrence of the network negatively correlated with SOVD task performance. Based on these methodological, behavioral, and neural findings in SOVD, in the second part of my thesis, I tried to experimentally induce specific misattributions of self-towards-other voices, following the prominent theory linking AVH with self-monitoring deficits. In order to do so, I combined the aforementioned SOVD task with a robotic procedure able to engender mild hallucinations in healthy individuals by perturbing bodily self-monitoring mechanisms. With such a procedure, I managed to alter self-voice perception of healthy participants (Study 3), what I further related to breathing (Study 4). Finally, I managed to induce identity-specific AVH in healthy individuals (Study 5), as quantified by the false alarm rate in a voice detection task, thereby contributing to the understanding of the mechanisms underlying AVH and relating them to impairments in bodily self-monitoring. Laying at the intersection of three sub-fields of neuroscience - voice perception, self-processing, and psychiatry - my thesis made a contribution by identifying neural correlates of SOVD, by proposing a new, multimodal perspective on self-voice-related research questions, and by demonstrating an experimental method of inducing AVH in a controlled laboratory environment. Together, my findings shed new light on the interactions between self-voice perception, sensorimotor processing, and interoception.

EPFL2020

Dissecting self-voice perception: From bone conduction to robotically-induced self-other voice misattribution in healthy listeners

Pavo Orepic

EPFL2020

Person Authentication by Fusing Face and Speech Information

Some applications of a priori knowledge in multi-stream HMM and HMM/ANN based ASR

Dissecting self-voice perception: From bone conduction to robotically-induced self-other voice misattribution in healthy listeners

Dissecting self-voice perception: From bone conduction to robotically-induced self-other voice misattribution in healthy listeners

Person Authentication by Fusing Face and Speech Information

Some applications of a priori knowledge in multi-stream HMM and HMM/ANN based ASR