Method and device for eliminating echo

Abstract

The invention discloses a method and a device for eliminating echo. The method comprises the steps of: determining the state of each subband adaptive filter respectively according to the cross correlation coefficient of an input signal on the near end of each subband and an echo estimation signal; when one end of the subband adaptive filter is in a speaking state, replacing the residual signal ofthe subband adaptive filter with comfortable noise of the subband, and outputting the replaced signal; and when both ends of the subband adaptive filter are in the speaking state, outputting the residual signal of the subband adaptive filter. In the invention, the signal processing characteristic of the subband can be fully utilized so that the residual echo can be suppressed more effectively. Inaddition, the suppression on the residual echo while both ends are in the speaking state and the protection on the voice of a person speaking on one end are enhanced, and the integral effect and the fluency of the system are improved.

Description

technical field [0001] The present invention relates to the technical field of communication, in particular to an echo cancellation method and a device thereof. Background technique [0002] In the voice communication system, after the remote input signal arrives at the local signal receiving device (such as a telephone), it passes through the speaker and room of the local signal receiving device to reach the earpiece. echo. In order to eliminate the echo, echo cancellation technology is needed to monitor the echo signal and eliminate the echo signal from the voice signal. For a long time, echo cancellation has been a very challenging task, mainly due to: [0003] (1) The acoustic echo will enter the microphone directly or after one or more reflections in a superimposed form, resulting in a very long tail of the echo and a long impulse response of the corresponding echo channel, generally hundreds of milliseconds. [0004] (2) The sound spectrum of the speech signal is no...

AI Technical Summary

Problems solved by technology

[0004] (2) The sound spectrum of the speech signal is non-flat and diffuse, and the commonly used adaptive algorithm is related to the statistical characteristics of the input signal. The diffusion of the eigenvalues ​​in the autocorrelation matrix of the speech signal will make, for example, NLMS (Normalized Least Mean Square, The adaptive convergence process of the normalized least mean square) algorithm slows down

[0021] (3) In some actual environments, due to the serious nonlinearity of the equipment or the serious environmental noise, the adaptive filter cannot converge well, resulting in obvious residual echo, and when there is an echo amplitude similar to the When the voice of a single-ended speaker is used, it is difficult for the nonlinear processing algorithm to distinguish the state of single-ended speaking and double-ended speaking. Misjudgment will lead to poor suppression of residual echo in single-ended speaking or serious sound cut-off in double-ended speaking

[0024] In the traditional system, the sub-band adaptive filtering output signal is synthesized into a full-band signal, and then processed by a post-processing algorithm based on the full-band. Such a post-processing algorithm does not make full use of the advantages of the sub-band to further improve the post-processing effect.

Images