Investigating the neural code for dynamic speech and the effect of signal degradation

Steadman, Mark (2015) Investigating the neural code for dynamic speech and the effect of signal degradation. PhD thesis, University of Nottingham.

PDF (Thesis - as examined) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (8MB) | Preview


It is common practice in psychophysical studies to investigate speech processing by manipulating or reducing spectral and temporal information in the input signal. Such investigations, along with the often surprising performance of modern cochlear implants, have highlighted the robustness of the auditory system to severe degradations and suggest that the ability to discriminate speech sounds is fundamentally limited by the complexity of the input signal. It is not clear, however, how and to what extent this is underpinned by neural processing mechanisms.

This thesis examines the effect on the neural representation of reducing spectral and temporal information in the signal. A stimulus set from an existing psychophysical study was emulated, comprising a set of 16 vowel-consonant-vowel phoneme sequences (VCVs) each produced by multiple talkers, which were parametrically degraded using a noise-vocoder. Neuronal representations were simulated using a published computational model of the auditory nerve. Representations were also recorded in the inferior colliculus (IC) and auditory cortex (AC) of anaesthetised guinea pigs. Their discriminability was quantified using a novel neural classifier.

Commensurate with investigations using simple stimuli, high rate envelope modulations in complex signals are represented in the auditory nerve and midbrain. It is demonstrated here that representations of these features are efficacious in a closed-set speech recognition task where appropriate decoding mechanisms are available, yet do not appear to be accessible perceptually.

Optimal encoding windows for speech discrimination increase from of the order of 1 millisecond in the auditory nerve to 10s of milliseconds in the IC and the AC. Recent publications suggest that millisecond-precise neuronal activity is important for speech recognition. It is demonstrated here that the relevance of millisecond-precise responses in this context is highly dependent on the brain region, the nature of the speech recognition task and the complexity of the stimulus set.

Item Type: Thesis (University of Nottingham only) (PhD)
Supervisors: Sumner, C.
Mason, R.
Keywords: Auditory, hearing, neuroscience, neural coding, auditory nerve, inferior colliculus, auditory cortex, speech, consonants, electrophysiology, machine learning, sound
Subjects: P Language and literature > P Philology. Linguistics
Q Science > QP Physiology > QP1 Physiology (General) including influence of the environment
W Medicine and related subjects (NLM Classification) > WV Otolaryngology
Faculties/Schools: UK Campuses > Faculty of Medicine and Health Sciences > School of Life Sciences
Item ID: 28839
Depositing User: Steadman, Mark
Date Deposited: 08 Oct 2015 09:34
Last Modified: 13 Sep 2016 18:33

Actions (Archive Staff Only)

Edit View Edit View