Double-talk robust VSS-NLMS algorithm for under-modeling acoustic echo cancellation

Most of the adaptive algorithms used for acoustic echo cancellation (AEC) are designed assuming an exact modeling scenario (i.e., the acoustic echo path and the adaptive filter have the same length) and a single-talk context (i.e., the near-end speech is absent). In real-world AEC applications, the adaptive filter works most likely in an under-modeling situation, i.e., its length is smaller than the length of the acoustic impulse response, so that the under-modeling noise is present. Also, the double-talk case is almost inherent, so that a double-talk detector (DTD) is usually involved. Both aspects influence and limit the algorithm's performance. Taking into account these two practical issues, a double-talk robust variable step size normalized least-mean- square (VSS-NLMS) algorithm is proposed in this paper. This algorithm is nonparametric in the sense that it does not require any information about the acoustic environment, so that it is robust and easy to control in practice.