Development of a Background Noise Cancellation System Using Efficient Oversampled DFT Filter Banks

Slow convergence and high computational complexity are the main problems incorporating with use of conventional adaptive noise filtering when applied to reduce the background in speech communication. In this paper, an improved background noise canceller for speech signals is developed, the canceller is derived by inserting an efficient two fold oversampled filter banks in the conventional fullband model. Proposed system equation is formulated with few realistic assumptions to ease analyzing and deriving an optimum prototype filter. The proposed oversampled system offers a simplified structure that without employing cross-filters or gap filter banks reduces the aliasing level in the subbands, and hence decreases the residual noise at the system output. The issue of increasing the initial convergence rate is addressed, and the computational complexity of the system is analyzed. The performance under white and colored environments is evaluated in terms of mean square error performance. Remarkably fast initial convergence was obtained. Moreover an increase in the amount of noise reduction by approximately 5dB compared to fullband model was reachable under actual speech and background noise. In spite of the insertion of analysis/synthesis filter banks, the proposed noise canceller is still computationally efficient.