Dual-channel speech enhancement method and device

Abstract

The invention discloses a dual-channel speech enhancement method and device, wherein the method includes: inputting a first sound signal received from a first microphone and a second sound signal received from a second microphone to a delay-and-add beamformer to obtain N beams; determining that the beam corresponding to the direction of the sound to be enhanced is a target beam, and the target beam is one of the N beams; calculating the existence probability of the sound to be enhanced in the target beam according to the N beam calculation; The target beam is enhanced based on the presence probability. By generating N beams based on the received two-channel sound signal, and in the case of a clear target beam, based on the joint calculation of the target beam and other beams, it is determined that the desired voice exists in the target beam, thereby avoiding voice from other directions. Interference, to obtain a higher signal-to-noise ratio.

Description

technical field [0001] The invention relates to the technical field of speech signal processing, in particular to a dual-channel speech enhancement method and device. Background technique [0002] In reverberant and noisy environments, multi-channel systems are usually used to obtain high-quality voice communication. Compared with single-channel systems, the spatial filtering capabilities of multi-channel can suppress interfering signals from undesired directions, so significant performance improvements can be obtained , and the greater the number of channels, the stronger the interference suppression ability. However, in the case of spatially incoherent noise fields, beamforming alone cannot provide sufficient noise reduction, and post-filtering needs to be introduced. [0003] The currently proposed post-filtering technology includes single-channel post-filtering. The single-channel post-filtering algorithm can better eliminate steady-state noise and is relatively stable w...

AI Technical Summary

Problems solved by technology

[0005] However, when this method is used in dual-channel speech enhancement, there will be a big problem, because the spatial filtering capability of the dual-channel is very limited, and the estimation of the blocking matrix will often have a large deviation, resulting in more noise in the reference channel. If the voice is expected to leak in, there will be a large error in the estimation of TBRR, which will lead to a serious self-elimination problem, which will seriously distort the voice, which is not conducive to subsequent interactive functions such as recognition.

Images