Stochastic filtering and speech enhancement using a recurrent quantum neural network
Laxmidhar Behera, Bharat Sundaram · 2004
This paper concerns with intelligent stochastic filtering using the recurrent quantum neural network model. The approach does not make any assumption about the nature and shape of both signal and noise. The recurrent quantum neural network (RQNN) is designed to model the unified response of a neural lattice while ignoring the individual neuronal responses. The average response of a neural lattice is described using Schrodinger wave equation. It is found that the closed loop RQNN dynamics exhibits soliton property. We have exploited this property for stochastic filtering in a recent communication. We further test the RQNN model for stochastic filtering of non-stationary signals which are characterized by time varying probability distribution functions (pdf). The performance efficacy of the RQNN in tracking amplitude modulated sinusoids and square waves signals is verified. The tracking of a speech signal embedded in Gaussian noise with non-stationary variance is also presented. The speech enhancement capability of the RQNN model is also tested in real-time.