【摘要】 提出了一种基于相位补偿改进的低信噪比多带谱减算法,构造了相位补偿函数对带噪语音相位谱进行补偿,使其更接近纯净语音相位,提升了多带谱减法在低信噪比下的性能.在Matlab仿真平台上就输出信噪比、PESQ和语谱图进行评估,所提算法均优于多带谱减算法,其中在-10 dB的极低信噪比下,该算法在PESQ上比原算法提高了26.4%.更多还原

【Abstract】 An improved low signal-to-noise ratio(SNR) multiband spectral subtraction algorithm is proposed based on phase compensation, constructed a phase compensation function to compensate the phase spectrum of noisy speech in order to make it closer to the phase spectrum of clean speech, and improved the performance of multiband spectral subtraction in low SNR. On the Matlab platform, the proposed algorithm is evaluated and has better performance than the multiband spectral subtraction algorithm in the output SNR, perceptual evaluation of speech quality(PESQ) and spectrogram. In the extremely low SNR of-10 dB, the proposed algorithm is 26.4% higher than the original algorithm in PESQ.更多还原