Echo removal

摘要

Echo removal techniques are described. As part of the echo removal, an adaptive model estimate of the echo in a received audio signal is determined using an adaptive model based on an outputted audio signal and the received audio signal. The adaptive model executes an algorithm comprising a convergence parameter to determine filter coefficients and uses said filter coefficients to filter the outputted audio signal to determine the adaptive model estimate of the echo. An accuracy value of the adaptive model is determined according to an echo return loss enhancement metric. The convergence parameter is updated based on the accuracy value. The adaptive model estimate of the echo is used to remove the echo in the received audio signal.

主权项

1. A method of removing echo, the method comprising:
outputting an output audio signal; receiving an input audio signal, wherein the received input audio signal includes echo resulting from said output audio signal; determining a first adaptive model estimate of the echo in the received input audio signal based on the outputted output audio signal and the received input audio signal; determining a performance value of the first adaptive model according to a performance metric; comparing the performance value with a threshold value; determining a second adaptive model estimate of the echo in the received input audio signal based on the outputted output audio signal, the received input audio signal, and further based on the comparison of the performance value with the threshold value; determining, based on the comparison of the performance value with the threshold value and further based on a relation between echo in the received input audio signal and the outputted output audio signal, whether an echo path is deemed linear; and in an event that the echo path is deemed linear:
using the first model estimate of the echo to remove the echo in the received input audio signal and updating a convergence parameter to control an adaptation speed of the first adaptive model; and in an event that the echo path is not deemed linear:
using the second model estimate of the echo to remove the echo in the received input audio signal and updating the convergence parameter to control the adaptation speed of the first adaptive model.