On detection of double talk and changes in the echo path using a Markov modulated channel model

This paper addresses the problem of detection and discrimination of the two phenomena, double talk and echo path change, in a telephone channel. It is of uttermost importance to quickly detect a change in the echo path while not confusing it with double talk, since the echo canceller should react differently whether an echo path change or double talk has occurred. Novel algorithms of low complexity are proposed. The system is described with a Markov modulated finite impulse response (FIR) filter. Depending on whether double talk or an echo path change occurs different parameters in the channel model change abruptly. Based on model assumptions, maximum likelihood (ML) parameter estimates of the communication channel are obtained via recursive (off-line) or iterative (on-line) methods using the expectation maximization (EM) algorithm. This enables us to use a hidden Markov model (HMM) state estimator to yield the minimum probability of error in identifying the state of the communication channel, i.e. the possible presence of double talk and/or echo path change. The proposed algorithms are experimentally verified using a real speech signal and impulse responses created from measured impulse responses from real hybrids