Echo Cancellation Method with Variable Step Size Based on Fast Convergence

Abstract

The invention discloses a variable-step-size echo cancellation method based on the rapid convergence characteristic. The method comprises the steps that Fourier transformation is carried out on a far-end reference signal x of a current frame and a microphone received signal d so as to acquire a corresponding far-end reference frequency domain signal X(k) and a microphone received frequency domainsignal D(k); according to the far-end reference frequency domain signal X(k) and the microphone received frequency domain signal D(k), a prior error E(k) is calculated; a transmission parameter T'i ofa filter is calculated; an updated step size mu(k) of the filter is calculated, and the filter is updated according to the updated step size mu(k); the current frame is calculated again according toupdated filter coefficients so as to acquire a posterior error Ep(k); and inverse Fourier transformation is carried out on the posterior error Ep(k), then the data of the latter half part is acquiredso as to acquire a time-domain signal of the current frame, and the step (A) is repeated to process the next frame. The variable-step-size echo cancellation method based on the rapid convergence characteristic has a good application prospect in a real-time voice dialogue system and in a scene with relatively high real-time performance requirements.

Description

technical field [0001] The invention relates to the technical field of acoustic echo cancellation, in particular to a variable step-length echo cancellation method based on fast convergence characteristics. Background technique [0002] With the rapid development of intelligent voice technology, more and more intelligent hardware provides the function of voice interaction, and the performance of acoustic echo cancellation is the main index to detect a voice interaction system. At present, the design of echo cancellation is mainly to add an echo cancellation chip to the hardware system. On the one hand, this echo cancellation chip obtains the voice input of the system from the microphone; on the other hand, it obtains the signal output from the hardware control unit to the speaker as a reference signal. Usually the microphone signal is a signal formed by mixing the voice command issued by the user and the sound broadcast by the smart hardware itself through the speaker, and t...

AI Technical Summary

Problems solved by technology

Moreover, this method does not have too many parameter configurations, but its tracking ability of path changes and steady-state misalignment is a particularly prominent contradiction

In the FDAF model, the larger the value of the system transfer parameter, the smaller the system's steady-state imbalance, but the filter tracking ability deteriorates significantly when the path changes, and even the filter freezes. Tracks mutation paths well, but increases homeostatic misalignment

Images