Research For And Evaluation On The Performance Of Speech Enhancement Algorithm

In this thesis, it discusses four types speech enhancement algorithms of different principles, which includes the wiener filtering method based on speech production model, the minimum mean square error method based on statistical model, based on signal subspace method, and the wavelet transform method. It describes these algorithms in details, and also introduces implementation steps. It is through simulation of each algorithm to get the enhanced speech signal. At the same time, it researches speech quality assessment algorithm. It carries out simulation experiments of segmental signal to noise ratio and perceptual evaluation of speech quality. It shows the performance evaluation results of the four types speech enhancement algorithm in this thesis.The main works of this thesis are as follows:First, four types of speech enhancement algorithms are the wiener filtering method, the minimum mean square error method, the signal subspace method, the wavelet transform method. They are simulated at the environment of white Gaussian noise, pink noise, factory noise. It can get enhanced speech signals through simulations.In this thesis, it compares the waveform's similarity of pure speech signal with enhanced speech signal in time domain. It can compare the signal to noise ratio of non-enhanced speech with enhanced speech.Second, the performance of speech enhancement algorithms ultimately depends on the quality of enhanced speech signal. In this thesis, it uses segmental signal to noise ratio to evaluate the speech enhancement algorithm for noise reduction level and the distortion performance. It gets the results of this performance evaluation.Third, it adopts the perceptual evaluation speech quality measure to evaluate speech intelligibility of enhancement algorithm. It gets the results of this performance evaluation. The correlation of traditional signal to noise ratio with subjective evaluation is very low and traditional signal to noise ratio doesn't reflect the speech intelligibility. On the other hand, the perceptual evaluation of speech quality algorithm is the best algorithm associated with the subjective evaluation. It can evaluate speech intelligibility accurately, and objectively. It has more consistent with the human ear perception.