Microphone array structure able to reduce noise and improve speech quality and method thereof

摘要

The present invention discloses a microphone array structure able to reduce noise and improve speech quality and a method thereof. The method of the present invention comprises steps: using at least two microphone to receive at least two microphone signals each containing a noise signal and a speech signal; using FFT modules to transform the microphone signals into frequency-domain signals; calculating an included angle between a speech signal and a noise signal of the microphone signal, and selecting a phase difference estimation algorithm, a noise reduction algorithm or both to reduce noise according to the included angle; if the phase difference estimation algorithm is used, calculating phase difference of the microphone signals to obtain a time-space domain mask signal; and multiplying the mask signal and the average of the microphone signals to obtain the speech signals of the microphone signals. Thereby is eliminated noise and improve speech quality.

主权项

1. A microphone array structure able to reduce noise and improve speech quality, comprising:
at least two microphones respectively receiving at least two microphone signals each containing a noise signal and a speech signal; at least two FFT (Fast Fourier Transform) modules transforming said microphone signals into frequency-domain signals; a processing unit calculating an included angle between said noise signal and said speech signal of said microphone signals and, selectively executing a spatial noise masking including a combination of a phase difference estimation with a masking estimation responsive to a non-zero value of said included angle, and executing a noise reduction to reduce noise responsive to a zero value of said included angle; a phase difference estimation module calculating phase difference and interaural time difference (ITD) of said microphone signals and identifying optimized ITD thresholds corresponding to said included angles, said thresholds are identified with a GSS (Golden Section Search) module; a mask estimation module using said thresholds to obtain a mask signal according to a binary mask, and multiplying said mask signal and an average of said microphone signals to obtain said speech signal of said microphone signal; and an IFFT (inverse-FFT)-OLA (overlap-and-add) module transforming said frequency-domain signals into time-domain signals; wherein said GSS module selects two points from a continuous range; said GSS module then compares function values of said two points and decreases size of said continuous range; and said GSS module then selects two additional points and compares function values thereof to continue decreasing size of said continuous range until a minimum function value is identified in said continuous range.