Flexible cochlear system based on digital model of cochlea: Structure, algorithms and testing

This paper presents complete speech processing system for cochlear prosthesis. The main elements of the system are speech enhancement system, speech processor and system for speech reconstruction. Speech enhancement system exploits properties of modulation spectrum of the human speech. Speech signal is split into 64 equal bands using maximally decimated polyphase DFT filter bank. Power spectrum of the signal for each channel is computed and filtered with tunable bandpass filter. Characteristic of the bandpass filter is computed independently for each channel base on the properties of power spectrum of the signal in the channel. Output signal is reconstructed from filtered power spectrum and original phase of the input signal. Tests proved that the system is suitable for cochlear implants and as a preprocessing procedure in Automatic Speech Recognition (ASR) systems. Speech processor dedicated for CIS stimulation was design base on digital, two dimensional, nonlinear model of the human cochlea model. The analysis filter bank was built with use tunable bandpass filters.