Network model training method, echo cancellation method and equipment

Abstract

The invention discloses a recurrent neural network model training method, an echo cancellation method, terminal equipment and a computer readable storage medium. The method comprises the steps: obtaining a sample signal which comprises a microphone signal and a reference signal corresponding to the microphone signal; carrying out feature extraction on the sample signal to obtain a first acoustic feature and a second acoustic feature, wherein the first acoustic feature is the acoustic feature of the microphone signal, and the second acoustic feature is the acoustic feature of the reference signal; inputting the first acoustic feature and the second acoustic feature into an initial recurrent neural network model, and outputting a predicted value of a phase sensitive mask, wherein the phase sensitive mask is a frequency spectrum proportion of an echo cancellation signal in the microphone signal, and a frequency spectrum comprises an amplitude spectrum and a phase spectrum; and calculating loss of the phase sensitive mask to adjust parameters of the initial recurrent neural network model to obtain the recurrent neural network model. In this way, the echo cancellation capability can be improved.

Description

technical field [0001] The present application relates to the technical field of echo cancellation, in particular to a training method of a cyclic neural network model, an echo cancellation method, a terminal device and a computer-readable storage medium. Background technique [0002] In video conferencing systems, due to the high-level coupling of microphones and speakers, acoustic echoes that affect voice interactions often occur. In traditional echo cancellation, the acoustic path from the speaker to the microphone in the acoustic environment is estimated through a finite-length linear filter to eliminate the linear echo, and then a gain value is applied to remove the remaining residual echo. However, in scenarios with low signal-to-return ratio and strong nonlinear distortion, the traditional echo cancellation algorithm cannot accurately estimate the residual echo, resulting in the inability of the video conferencing system to suppress the residual echo and cause near-en...

AI Technical Summary

Problems solved by technology

However, in scenarios with low signal-to-return ratio and strong nonlinear distortion, the traditional echo cancellation algorithm cannot accurately estimate the residual echo, which leads to the inability of the video conferencing system to suppress the residual echo and cause near-end voice distortion

Images