"Optimum" filter for speech enhancement using integrated digital signal processors

An algorithm for adaptive filtering of speech in additive acoustic noise is presented. In comparison with other speech enhancement algorithms a very low computational cost is required. The efficiency of the algorithm is achieved by proper modifications of a Wiener optimum filter with a filter bank. Due to the use of digital Lerner band pass filters a very compact filter bank structure is obtained. The speech enhancement system has been implemented with INTEL 2920 signal processors. As a result, the quality of noisy speech signals band-limited to (300-3.300) Hz can be improved for input signal to noise ratios as low as -5dB.