语音增强.pptVIP

basic spectral subtraction You can hear musical noise obviously.It has great influence on listening. basic Wiener filter Iterating wiener filter Iterating wiener filter Iterating wiener filter is the priori error estimate of , is the posteriori error estimate of . priori error estimate and posteriori error estimate is： Covariance： improved subtraction Iterating wiener filter kalman filter Firstly, Spectral subtraction has great effect on white noise ,and improved method suppress the musical noise. But its not good at narrow-band noise. Secondly, Wiener filter has almost the same effect on the noise. Thirdly, Kalman filter has overcome the shortage of Wiener filter that so many previous datas are need ,so it has been used in more and more fields.However,we didnt make it in Kalman filtering and show its advantage. Kalman filter Kalman filter Recurrence formula K is the correction matrix，also known as Blend factor. Kalman filter The illustrative diagram Kalman filter is a data processing algorithm of optimal recursive. optimal most efficient most useful to solve most problems filtered speech The noise is still mixed up with the speech.KF should be the near optimal algorithm theoretically.Its probably owing to the program. As everyone can see,at the front part of the program,the noise all we use is additive white Gaussian noise.Now,we will use a speech segment which mixed with train noise to test. noisy speech clean speech filtered speech spectrogram of noisy speech spectrogram of clean speech spectrogram of filtered speech filtered speech filtered speech Enhance-ment /SNR Basic spectral subtraction Improved Spectral Subtraction Basic Wiener filter Iterating Wiener filter Kalman filter noisy speech 12.2826 12.2826 12.2826 12.2826 12.2826 filtered speech 18.3314 17.0264 18.3314 14.5978 7.4447 improve-ment 6.0498 4.7438 6.0498 2.3152 -4.