A new doubletalk detector with fuzzy decision thresholding

A doubletalk detector (DTD) is used with an echo canceller to sense when far end speech is corrupted by near end speech. This detection is used to stop the divergence of an adaptive echo canceller. In the steady state, this decision is crucial because it controls the operation of the nonlinear processor (NLP) used for the suppression of residual echo. Various algorithms have been proposed in the literature, all of them having a crisp decision threshold. We review existing algorithms and explore their performance and suitability for real time fixed-point implementation. A new variance based normalized estimator for doubletalk detection is proposed. Finally, a fuzzy system is designed to select the threshold for the DTD during operation. This new algorithm takes into account the near end to far end power ratio, is simple to implement and shows an enhancement in performance. Furthermore, precision requirements are not very stringent for the fuzzy DTD, thus making it suitable for fixed-point implementation.