Coherence based double talk detector with soft decision

Acoustic echo cancellation is one of the oldest applications of adaptive filters and today is a part of each speakerphone. An important block of each acoustic echo canceller is the double talk detector. It blocks the adaptation of the filter when near end voice is present and thus preventing the adaptive filter from diverging from the optimal position. In this paper we present an improved version of coherence based double talk detector. It provides estimation of the double talk presence probability per bin and per frame and has better precision compared to the baseline algorithm.