Robust GSC method based on coherence and energy ratio

Abstract

The invention relates to a robust GSC method based on coherence and an energy ratio, and belongs to the technical field of microphone array speech enhancement, and the method comprises the following steps: S1, outputting an expected signal at an upper branch part by adopting a super-directional fixed beamformer, and solving the coherence of the expected signal and an original signal; S2, using a delay summation fixed beamformer to output a signal, and calculating a signal to interference energy ratio with an original signal in a certain frequency range; and S3, jointly controlling updating ofthe adaptive noise canceller ANC by using the solved coherence and energy ratio. Compared with a traditional GSC algorithm, the scheme has better low-frequency characteristics and beam directivity, ANC updating is jointly controlled through coherence and energy ratio, the problem that the ANC mistakenly eliminates expected signals in a voice segment is prevented, and higher robustness is achieved.

Description

technical field [0001] The invention belongs to the technical field of microphone array speech enhancement, and relates to a robust GSC method based on coherence and energy ratio. Background technique [0002] In recent years, with the rapid development of artificial intelligence, speech as an important entrance of human-computer interaction has received widespread attention, and the research on speech interaction at home and abroad has set off a new climax. From Amazon's smart speaker Echo to Xiaomi's smart speaker Xiao Ai, to Baidu's Xiaodu Xiaodu, these voice interaction products have gradually integrated into our lives, and the front-end noise reduction mainly uses microphone arrays to suppress noise. To achieve the purpose of speech enhancement. The microphone array utilizes the spatial information of the speech signal, which has more space-time information than the single-channel speech signal, and has the advantages of higher gain, beam flexibility, and strong noise ...

AI Technical Summary

Problems solved by technology

In the traditional GSC algorithm, due to the leakage of the expected signal in the lower branch, the mismatch of the microphone, the reverberation in the room, and the inaccurate estimation of the speech segment, it is easy to cause the false cancellation of the expected signal. In order to solve this problem, Sven et al. Update the adaptive noise canceller when the interference noise is dominant, Hoshuyama proposed to use the adaptive blocking matrix output to constrain the update of the filter, Cohen et al proposed the acoustic transfer function from the source signal to the microphone array to avoid the desired signal cancellation, but there are still deficiencies

Images