Design And Implementation Of Hybrid Digital Filter System Based On Fuzzy Inference

At present,there are many means and devices for electronic communication in society,but voice communication is an important link to maintain the emotion between people and people,and it is in an important position.The voice call process may be in a noisy environment contaminated by noise,affecting the quality of communication.Therefore,it is necessary to use a noise suppression algorithm to develop a voice enhancement system to reduce the negative effects of noise.For the characteristics of unsteady type noise,the traditional filtering algorithm has insufficient noise filtering,or it filters out parts of clean speech,resulting in shortcomings such as speech distortion.Since the noise will affect the quality of the voice recording,the acoustic model of the training does not match the actual application environment,so that the recognition accuracy of the voice recognition system is greatly reduced.In this paper,we especially analyze and study the damage of speech characteristics caused by additive noise,and deal with the measured noise to expect better speech enhancement results.The research work in this paper includes several aspects: Firstly,the related theories such as spectrum subtraction method,wavelet transform and decomposition,Wiener filtering,fuzzy theory and speech enhancement are summarized.Secondly,the signal preprocessing and speech endpoint detection are designed and implemented.The key points are designed to sample the bits in the signal preprocessing process,pre-emphasis,framing,windowing,etc.The coding realizes the corresponding functions and processing effects.Voice endpoint detection method and its detection accuracy rate.Thirdly,the hybrid filtering speech enhancement system is implemented,and the fuzzy inference system architecture is designed.The fuzzy controller,speech feature extraction,signal type induction,fuzzy rule inference and training process are realized.The speech enhancement system uses a speech signal containing noise as an input signal,uses a speech noise prediction model,and establishes a model-based processing method to suppress noise outside the speech itself.The speech signal is subjected to both steady state and non-steady state type discrimination.The speech features are analyzed again,and six common features are used to identify whether the speech is a steady-state or non-steady-state signal.Finally,the hybrid filter voice enhancement system is tested,and the Aurora-voice database is used to provide the audio file content and obtain the sampling frequency.Using the signal-to-noise ratio SNR commentary sound quality method,the SNR value corresponding to the 8 kinds of noises included in the Aurora voice database and the SNR value of the enhanced speech signal are obtained,and various enhanced filtering results and three filter combinations are obtained.the result of.Through the research of this paper,an effective speech enhancement method is proposed.The characteristics of the model signal in the input speech signal are combined,and the spectrum subtraction method is combined with the wavelet filtering and Wiener filtering three filtering methods to emphasize the fuzzy inference system.The distribution ratio of the above three filters restores the ideal speech signal,minimizes the influence of ambient noise and speech,and compensates for the shortcomings of using a single filtering method,thereby effectively improving the SNR value.Therefore,the research in this paper has very important application value.