A real-time implementation of the primary auditory neuron activities

Computational models of the auditory pathway simulate the different stages of the auditory periphery, which includes the outer, middle and inner ear stages. The studies of the levels of the auditory pathway beyond the inner ear stages require the availability of data primarily of the cochlea response. If the computational model of the auditory pathway simulates the cochlea responses slowly, the responses of the higher levels of the auditory pathway will also be slow. Hence, a real-time computational model of the auditory pathway provides the capability to study higher levels of sound perception studies in the field of computational neuroscience by providing on-the-fly or immediate responses of the cochlea within the auditory pathway. In this thesis, the development of a real-time computational model of an auditory pathway is discussed. A review of five auditory pathway computer models is presented and a model is selected for implementation into a real-time computer model. The transition from the original model to a real-time implementation includes a translation to C language before being integrated with JUCE, a C++ graphical user interface library. The input signals in the real-time model are generated either through a software sine tone generator or acquired from a microphone channel on a computer. As part of cochlea simulation, the algorithms are divided to generate responses in channels. A large number of channels results in a finer resolution of spectral projection of the cochlea response. To achieve the optimum number of channels in real-time, POSIX threads are used to achieve computing parallelism. As part of optimisation to load more channels, mathematical optimisation is studied and utilised in the real-time model. It will be demonstrated in this thesis that the RMS errors of the responses of the developed real-time computer model of the auditory pathway as opposed to the original model measures below 1% and its maximum load is dependent on the computer it runs on. On a laptop with a dual core CPU, the real-time model is able to simulate 85 channels of the basilar membrane displacement whereas a desktop with a quicker dual core CPU model accommodates twice as many channels. With math optimisation enabled, there is 13% and 8% increase in the computation of channels for the laptop and desktop respectively. However, the RMS errors of the real-time model with math optimisation enabled and the original model increases to 8% due to approximation errors