本文嘗試將原始語音嵌入零值音框，再使用雜訊頻譜峰值鎖定的方式估測雜訊，提升語音音框中雜訊頻譜估測的位準，使得增強後的語音，可以有效的移除背景雜訊及改善殘留雜訊過多的缺點。零值音框嵌入法是在語音信號中，嵌入零值音框，但是在通道中將受到通道雜訊的干擾，導致語音品質下降。由於零值嵌入處沒有語音信號，因此接收端可以利用該處估測雜訊頻譜強度；我們在零值嵌入處估測雜訊頻譜強度，並且施予極大值鎖定，亦即選取雜訊頻譜強度的累積最大值，當作估測值，使得估測雜訊強度不會偏低，確保語音增強的效能。實驗結果證明：本文所提出的方法，可以大幅改善零值音框嵌入法的雜訊移除效能，而且在增強語音中的殘留雜訊也大幅降低。 This paper proposesincreasing theestimated magnitude ofnoise spectrum in a zero-padded frameby peak-spectrum-holding method, enabling the background noise to be efficiently removed in enhanced speech; meanwhile, the quantity of residual noise is also reduced.The idea of frame-zero-padding method is to periodically insert zero frames into a clean speech signal. Hence this zero padded signal is transmittedto the receiver through anadditive noisechannel. The quality of transmitted speech is deteriorated. Since speech is absent in the zero-padded frames, we can estimate the magnitude of noise spectraduringtheseframesin the receiver. In order to improve the performance of removing background noise, we propose holdingthe magnitude of peak spectrum in zero-paddedframes. This peak magnitude is kept unchanged until meeting a stronger noise spectrum. Experimental results show that the proposed methodsignificantly outperforms the frame-zero-paddingmethodby removing more amountsof background noiseand residual noise.
