Signal processing method for enhancing target voice signal pickup in sound environment

Abstract

The invention relates to a signal processing method for enhancing target voice signal pickup in a sound environment. The method includes the steps that (1), parameters of an ESN are acquired through experiments, and a corresponding sound source model is established; (2), the model is used for two occasions, namely, when output of the model is certain expected target voice signals and input of the model is a mixture of sound environment reflected sound signals of a target voice source and the target voice signals, the model can be used for echo cancellation of field amplified sound; when the output of the model is the expected target voice signals and the input of the model is a mixture of sound environment reflected sound signals of another specific voice source and the target voice signals, the model can be used for echo cancellation of voice communication between two specific persons; (3), when the model is used by a target voice person in the actual sound environment and the pickup position changes, reflected signals of sound source signals pointed by training can be suppressed as well, and then corresponding enhanced target voice signals are output. By the adoption of the signal processing method, the influence of the change of the pickup position on quality of the voice signals is overcome.

Description

technical field [0001] The invention belongs to the processing technology of indoor voice signal picking, and relates to a digital signal processing method for parameter selection and training modeling of an echo state neural network through experiments, in particular to a signal processing method for picking up target voice signals in an enhanced acoustic environment. Background technique [0002] In live sound reinforcement, the objects involved in eliminating the influence of echo are: a specific target voice and the ambient reflection sound of the specific target voice, which is mainly used to increase the sound gain. The main related technologies are: (1) Traditional technologies such as narrow-band equalization are the processing of filtering peaks and eliminating feedback natural vibration; A common problem that exists is that the processing technology is complex, and it is not conducive to the fidelity of the speech signal; (2) The method based on modern digital sign...

AI Technical Summary

Problems solved by technology

(2) Algorithms based on adaptive filtering The commonly used LMS algorithm has a small storage capacity, is easy to implement and detect, but has poor convergence; and the RLS algorithm with good convergence has a large amount of calculation, so there are many improved algorithms for them, and applied to Adaptive Filtering and Cancellation Algorithm Processing for Solving Practical Echo Problems

If the two signals passed to the echo cancellation algorithm are not well synchronized, that is, the frame misalignment of the two signals occurs, it will be difficult to cancel

[0005]Secondly, the acoustic echo adaptive filter elimination technology formed by the acoustic coupling between the speaker and the microphone has the following technical problems: (1) Due to the long delay time (up to 1s), requires a high-order filter with thousands of coefficients to fit, requiring more computing resources

(2) It is difficult to stabilize such a long high-order filter and improve its adaptive speed

Firstly, the acoustic echo path is not stable due to the change of acoustic characteristics; secondly, the acoustic echo is propagated through multiple channels; thirdly, the propagation and scattering characteristics of the room acoustic space are nonlinear, and the general (or ) linear filters do not perform well on modeling

Images