Method and device for eliminating residual echo and noise

Abstract

Embodiments of the present application disclose a method and device for eliminating residual echo and noise. The method includes: framing, windowing, and Fourier transforming a received voice time-domain signal containing echo and noise and a far-end reference acoustic time-domain signal Transform to obtain the corresponding frequency domain signal and determine the echo frequency domain signal, and further determine the voice and audio domain signal containing residual echo and noise; The respective amplitude spectra of the signals are energy normalized to obtain the corresponding features; according to the corresponding features and the cascaded network after training, the target voice and audio domain signals are determined; the target voice and audio domain signals are subjected to inverse Fourier transform to obtain the target Speech time domain signal. The embodiment of the present application uses a feature attention model to assign different importance to input features, so as to reduce redundant information in the input features. The multi-domain loss function is used to train the cascade network to reduce the sensitivity of the model to signal energy.

Description

technical field [0001] The invention relates to the field of echo and noise elimination. In particular, it relates to a residual echo and noise elimination method and device. Background technique [0002] At present, the echo cancellation technology is mainly to remove the echo signal formed by the far-end reference sound signal in the speech signal, and the speech noise reduction technology is mainly to remove the background noise and directional noise interference in the speech signal. Both echo cancellation technology and speech noise reduction technology are designed to improve the quality and intelligibility of speech. In the echo cancellation technology, combining the adaptive filtering method based on traditional signal processing and the residual echo cancellation method based on deep learning can effectively improve the generalization performance of the system. [0003] However, in traditional methods, residual echo and noise cancellation are often performed indep...

AI Technical Summary

Problems solved by technology

When training the residual echo and noise cancellation model, most of the existing technologies use the mean square error of the target amplitude spectrum and the estimated amplitude spectrum as the loss function, but the above loss function depends on the energy of the signal, and the scale of the signal with different energy is also different. will be different

Images