Double‐talk detection for adaptive noise canceller

Abstract

AbstractIt is known that the estimation accuracy deteriorates in the double‐talk period in the traditional adaptive noise canceller. The deterioration can be ameliorated by the conventional method if the double‐talk period can be detected. However, the detection of the double‐talk period has not been presented (a method is proposed for the adaptive echo canceller, which, however, has a problem in the direct application to the adaptive noise canceller).From such a viewpoint, this paper proposes a double‐talk detection method for the adaptive‐noise canceller. The method is based on the correlation of the exponential weighted square sum and has the following features. No parameter needs to be known beforehand. The result is less affected by the path variation. The distortion to the desired signal can be suppressed. The computational complexity can be reduced by decimation. The method can also be applied to the adaptive echo canceller.In this paper, the validity of the method is demonstrated by a theoretical analysis. The following result is obtained by a computer simulation. When the adaptive noise canceller is controlled by the proposed method, the accuracy of the estimation is improved by approximately 30 dB and the distortion to the desired signal is suppressed by approximately 15 dB, compared to the case where the method for the adaptive echo canceller is used.