Research On Wiener Filter Speech Enhancement Algorithms With Intelligibility Constraints

In today's society,the communication between people and machines is very frequent and close,so the level of speech signal processing is required to be higher.Because the main medium of speech signal transmission is air,the speech signal will inevitably be disturbed by the noise in the air in the transmission process,which will affect the correct reception of the speech signal at the receiving end.In this context,the research of speech enhancement algorithm has a strong practical significance.The purpose of speech enhancement is to eliminate noise to the greatest extent and improve the quality and intelligibility of enhanced speech.However,the improvement of speech quality does not necessarily lead to the improvement of intelligibility,which is due to the distortion of enhanced speech in the process of noise suppression.According to the ratio of enhanced speech to pure speech amplitude in frequency domain,the distortion of enhanced speech can be divided into three categories: attenuation distortion,amplification distortion with amplification factor less than 6 dB and amplification distortion with amplification factor greater than 6 dB.Existing research shows that amplification distortion with amplification multiple more than 6 dB has a great impact on speech intelligibility.Removing this distortion completely can significantly improve the intelligibility of enhanced speech.In view of the above problems,this paper first introduces the development history and basic theory of speech enhancement technology,and focuses on the principle of Wiener filtering algorithm.Then,in order to solve the amplification distortion of existing speech enhancement algorithms when the magnification factor exceeds 6 dB,we need to assume that the pure speech amplitude is known and that the pure speech and noise are perpendicular to each other.Then,by introducing the geometric phase relationship among noisy speech,pure speech and noise,this paper transforms the constraints of pure speech amplitude or prior signal-to-noise ratio in existing algorithms into the constraints of instantaneous prior signal-to-noise ratio and instantaneous posterior signal-to-noise ratio,and proposes a Wiener filter speech enhancement algorithm to solve the amplification distortion of amplification multiple more than 6 dB.Finally,on the basis of theoretical analysis and formula derivation,combined with simulation experiments,it is verified that the proposed algorithm is better than the existing algorithms in speech intelligibility and speech quality.