Double-microphone array echo eliminating method, device and electronic equipment
Embodiments of the present application relate to a dual-microphone array echo cancellation method, device and electronic equipment, comprises: the acquired distal signal, the first proximal signal and the second proximal signal are processed by linear filtering; obtain initial suppression gain factors and variable step-size factors of the first near-end signal and the second near-end signal; performing residual echo suppression on the first error frequency spectrum and the second error frequency spectrum by using an adaptive zero-pole echo canceller to obtain a target frequency spectrum signal; performing sub-band range selection on the initial suppression gain factors to obtain a smooth factor, and performing full-band smoothing and exponential operation to obtain a secondary suppression gain factor; performing filtering processing on the target frequency spectrum signal by using the secondary suppression gain factor to obtain a target near-end voice signal.
1. A dual-microphone array echo cancellation method, the method comprising:
acquiring far-end signals generated by an electronic equipment during a phone conversation, and a first near-end signal and a second near-end signal of a dual microphone array;
performing Fourier transform on the far-end signals to obtain far-end frequency domain information, performing Fourier transform on the first near-end signal to obtain first near-end frequency domain information, and performing Fourier transform on the second near-end signal to obtain second near-end frequency domain information;
performing filtering processing on the far-end frequency domain information by using the updated filter weight coefficient of the previous frame to obtain an echo frequency spectrum;
subtracting the echo frequency spectrum from the first near-end frequency domain information to obtain a first initial error frequency spectrum;
subtracting the echo frequency spectrum from the second near-end frequency domain information to obtain a second initial error frequency spectrum;
performing centroid fitting on the first initial error frequency spectrum and the second initial error frequency spectrum respectively to obtain initial suppression gain factors of the first near-end signal and the second near-end signal, which comprises calculating a first cross-correlation coefficient between the far-end frequency domain information and the first initial error frequency spectrum, and a second cross-correlation coefficient between the far-end frequency domain information and the second initial error frequency spectrum, performing centroid fitting on the first cross-correlation coefficient to obtain a first initial suppression gain factor, and performing centroid fitting on the second cross-correlation coefficient to obtain a second initial suppression gain factor, multiplying the first initial suppression gain factor by the first initial error frequency spectrum to obtain the first error frequency spectrum, and multiplying the second initial suppression gain factor by the second initial error frequency spectrum to obtain the second error frequency spectrum, and using the initial suppression gain factors as variable step-size factors for linear filtering processing of the next frame;
performing residual echo suppression on the first error frequency spectrum and the second error frequency spectrum by using an adaptive zero-pole echo canceller to obtain a target frequency spectrum signal;
after performing sub-band range selection on the initial suppression gain factors, obtaining a smooth factor, and performing full-band smoothing and exponential operation to obtain a secondary suppression gain factor;
performing filtering processing on the target frequency spectrum signal by using the secondary suppression gain factor to obtain a target near-end voice signal.
2. The method of claim 1 , the operation of using the initial suppression gain factors as variable step-size factors for linear filtering processing of the next frame comprises:
updating the filter weight coefficient of the previous frame based on the first initial suppression gain factor and the second initial suppression gain factor to obtain the filter weight coefficient of the current frame;
using the first initial suppression gain factor and the second initial suppression gain factor as variable step-size factors for linear filtering processing of the next frame, if the first initial suppression gain factor and the second initial suppression gain factor are smaller than a preset threshold.
3. The method of claim 1 , the operation of performing residual echo suppression on the first error frequency spectrum and the second error frequency spectrum by using an adaptive zero-pole echo canceller to obtain a target frequency spectrum signal comprises:
calculating a first branch frequency spectrum signal based on the first error frequency spectrum and a time delay compensation, and calculating a second branch frequency spectrum signal based on the second error frequency spectrum and the time delay compensation;
calculating a differential signal cross-correlation power spectrum coefficient between the first branch frequency spectrum signal and the second branch frequency spectrum signal and a weight coefficient of the adaptive zero-pole echo canceller based on the first branch frequency spectrum signal and the second branch frequency spectrum signal;
performing residual echo suppression on the first error frequency spectrum and the second error frequency spectrum by using the weight coefficient of the adaptive zero-pole echo canceller to obtain the target frequency spectrum signal.
4. The method of claim 1 , the operation of after performing sub-band range selection on the initial suppression gain factors, obtaining a smooth factor, and performing full-band smoothing and exponential operation to obtain a secondary suppression gain factor comprises:
performing sub-band range selection on the initial suppression gain factors to obtain the initial suppression gain factors in a plurality of sub-band ranges;
sorting the initial suppression gain factors in the plurality of sub-band ranges, and calculating a suppression gain coefficient smooth factor;
performing full-band smoothing processing and exponential operation on the initial suppression gain factor by using the suppression gain coefficient smooth factor to obtain the secondary suppression gain factor.
5. The method of claim 1 , the operation of performing filtering processing on the target frequency spectrum signal by using the secondary suppression gain factor to obtain a target near-end voice signal comprises:
performing noise reduction processing on the target frequency spectrum signal;
multiplying the secondary suppression gain factor by the target frequency spectrum signal after noise reduction, performing inverse Fourier transform, and performing overlap-adding to obtain the target near-end voice signal.
6. An electronic equipment, the electronic equipment comprising:
a dual microphone array, at least one processor, and a memory, wherein the memory is communicatively connected with the processor, and the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform:
acquiring far-end signals generated by an electronic equipment during a phone conversation, and a first near-end signal and a second near-end signal of a dual microphone array;
performing Fourier transform on the far-end signals to obtain far-end frequency domain information, performing Fourier transform on the first near-end signal to obtain first near-end frequency domain information, and performing Fourier transform on the second near-end signal to obtain second near-end frequency domain information;
performing filtering processing on the far-end frequency domain information by using the updated filter weight coefficient of the previous frame to obtain an echo frequency spectrum;
subtracting the echo frequency spectrum from the first near-end frequency domain information to obtain a first initial error frequency spectrum;
subtracting the echo frequency spectrum from the second near-end frequency domain information to obtain a second initial error frequency spectrum;
performing centroid fitting on the first initial error frequency spectrum and the second initial error frequency spectrum respectively to obtain initial suppression gain factors of the first near-end signal and the second near-end signal, which comprises calculating a first cross-correlation coefficient between the far-end frequency domain information and the first initial error frequency spectrum, and a second cross-correlation coefficient between the far-end frequency domain information and the second initial error frequency spectrum, performing centroid fitting on the first cross-correlation coefficient to obtain a first initial suppression gain factor, and performing centroid fitting on the second cross-correlation coefficient to obtain a second initial suppression gain factor, multiplying the first initial suppression gain factor by the first initial error frequency spectrum to obtain the first error frequency spectrum, and multiplying the second initial suppression gain factor by the second initial error frequency spectrum to obtain the second error frequency spectrum, and using the initial suppression gain factors as variable step-size factors for linear filtering processing of the next frame;
performing residual echo suppression on the first error frequency spectrum and the second error frequency spectrum by using an adaptive zero-pole echo canceller to obtain a target frequency spectrum signal;
after performing sub-band range selection on the initial suppression gain factors, obtaining a smooth factor, and performing full-band smoothing and exponential operation to obtain a secondary suppression gain factor;
performing filtering processing on the target frequency spectrum signal by using the secondary suppression gain factor to obtain a target near-end voice signal.
7. The electronic equipment of claim 6 , the dual microphone array is an end-firing arrangement.
8. A non-transitory computer readable storage medium, the computer readable storage medium storing computer-executable instructions which, when executed by an electronic equipment, cause the electronic equipment to perform:
acquiring far-end signals generated by an electronic equipment during a phone conversation, and a first near-end signal and a second near-end signal of a dual microphone array;
performing Fourier transform on the far-end signals to obtain far-end frequency domain information, performing Fourier transform on the first near-end signal to obtain first near-end frequency domain information, and performing Fourier transform on the second near-end signal to obtain second near-end frequency domain information;
performing filtering processing on the far-end frequency domain information by using the updated filter weight coefficient of the previous frame to obtain an echo frequency spectrum;
subtracting the echo frequency spectrum from the first near-end frequency domain information to obtain a first initial error frequency spectrum;
subtracting the echo frequency spectrum from the second near-end frequency domain information to obtain a second initial error frequency spectrum;
performing centroid fitting on the first initial error frequency spectrum and the second initial error frequency spectrum respectively to obtain initial suppression gain factors of the first near-end signal and the second near-end signal, which comprises calculating a first cross-correlation coefficient between the far-end frequency domain information and the first initial error frequency spectrum, and a second cross-correlation coefficient between the far-end frequency domain information and the second initial error frequency spectrum, performing centroid fitting on the first cross-correlation coefficient to obtain a first initial suppression gain factor, and performing centroid fitting on the second cross-correlation coefficient to obtain a second initial suppression gain factor, multiplying the first initial suppression gain factor by the first initial error frequency spectrum to obtain the first error frequency spectrum, and multiplying the second initial suppression gain factor by the second initial error frequency spectrum to obtain the second error frequency spectrum, and using the initial suppression gain factors as variable step-size factors for linear filtering processing of the next frame;
performing residual echo suppression on the first error frequency spectrum and the second error frequency spectrum by using an adaptive zero-pole echo canceller to obtain a target frequency spectrum signal;
after performing sub-band range selection on the initial suppression gain factors, obtaining a smooth factor, and performing full-band smoothing and exponential operation to obtain a secondary suppression gain factor;
performing filtering processing on the target frequency spectrum signal by using the secondary suppression gain factor to obtain a target near-end voice signal.