Method and device of echo elimination

Abstract

The invention discloses a method and a device of echo elimination. The method comprises the following steps of: determining the energy of remote input signals and residual signals by a self-adapting filter according to the remote input signals in the last filtering treatment and the residual signals obtained in the last filtering treatment; determining filter coefficients used in the current filtering treatment by the self-adapting filter according to the filter coefficients used in the last filtering treatment, the remote input signals, the residual signals and the energy sum of the remote input signals and the residual signals; and executing the current filtering treatment by the self-adapting filter according to the determined current filter coefficients. The invention realizes the echo elimination in voice communication under the condition that double-talk detection is not carried out.

Description

technical field [0001] The present invention relates to the communication field, in particular to an echo cancellation method and device. Background technique [0002] In a voice communication system, after the remote input signal reaches the local signal receiving device (such as a telephone), it passes through the local signal receiving device's speaker, room and other devices to reach the handset. Sound reflections create echoes. In order to eliminate the echo, in the prior art, an adaptive filter is used to perform filtering before outputting the speech signal, and a widely used filtering algorithm is adaptive filtering normalized least mean square algorithm (NLMS). [0003] The echo cancellation process based on the NLMS algorithm is as follows: figure 1 , where x represents the far-end input signal, y represents the actual echo signal formed by x passing through the room, v represents the voice of the local speaker and background noise, and d represents the near-end ...

AI Technical Summary

Problems solved by technology

[0030] However, the above-mentioned energy method and correlation calculation method all require that the adaptive filter is already in a convergent or close to convergent state, and it is often difficult to meet this condition in the actual application of echo cancellation. For example, when a conversation starts in a communication, a double The probability of sounding at both ends is higher than that during a call. At this time, the adaptive filter is still in the process of converging, so use the energy method and correlation calculation method to detect double-ended sounding, and control the adaptive filter to stop iterating, which will cause echo Eliminate the vicious cycle of performance degradation and misdetection

At the same time, the false detection in the double-talk detection will cause the adaptive filter coefficients to not be effectively updated, and may also cause a vicious cycle of echo cancellation performance deterioration and detection errors

Images