The cochlear implant has been implanted in thousands of people throughout the world. However, the benefit received from the implant varies considerably. Some people can obtain near perfect perception of speech while others struggle. This project will develop specialised signal processing techniques that will help to better distinguish the different speech sounds.

This research involves careful examination of speech perception by people using the implant. The results of these studies are used to design strategies which will provide the features to the cochlear more adequately and in a way that will improve their perception by the users.

Modelling

This project is investigating the effects of neural network activity in the processing of sounds by the cochlear nucleus, the first processing centre in the auditory system.  We aim to show that inhibitory inter-neurons enhance the input from the hearing nerve to higher auditory centres in the brain.  This mechanism is particularly useful in processing speech sounds.

As yet, a firm hypothesis has not been established as to how neurons in the cochlear nucleus utilise inhibitory inter-neurons to enhance processing.  Existing models of individual neuron characteristics are unable to explain some functional behaviours seen in animal studies.  A greater understanding of the network behaviour in the cochlear nucleus will contribute to our understanding of sound processing by the human auditory system, hence enable us to develop better strategies for cochlear implant processing.

Optimisation

Synaptic properties within microcircuit models are highly influenced by driven input and background noise sources; therefore optimization routines must be carefully designed to be robust to trial-to-trial variation and maintain parameter sensitivity of the cost function around the target data. This study presents a method for optimizing topographically ordered biological neural networks using genetic algorithms. Neural responses are generated with surrogate parameters and the ability of the algorithm to find the known neural parameters being investigated in a current study.