Research On Multi-level FrFT Speech Enhancement Algorithm Based On Sparse Metric

Speech has been the main carrier of daily communication between people.With the rapid development of artificial intelligence technology,speech has also become an important form of human-computer interaction.However,speech is interfered by the noise of the surrounding environment,which reduces the quality and intelligibility of the speech,and seriously affects the quality of people’s life and that of human-computer interaction.Speech enhancement is that some algorithms are used to eliminate the noise in the speech as much as possible and to improve the intelligibility and intelligibility of the speech.Traditional speech enhancement algorithms have a good noise reduction effect in the stable noise environment,but in a non-stationary noise environment,these traditional methods have no obvious noise reduction effect.Even the quality of the speech is reduced while removing the noise,which leads to the speech distortion seriously.The traditional transform domain speech enhancement algorithms are analyzed in this paper.In non-stationary environments,the overlapping speech signals and noise in the traditional transform domain cannot be completely separated.Based on this,the multi-stage FrFT speech enhancement algorithm based on sparsity is proposed.Research content and innovations are as follows:(1)Sparsity methods research.The energy distribution of the speech and noise in the fractional Fourier transform domain is studied,and it is found that the speech signal has a strong energy aggregation in the fractional domain.Traditional optimal order determination methods are fully studied,such as the Minimum Mean Square Error method and the maximum Signal-to-Noise Ratio method.These methods have the disadvantage of being computationally intensive and can not applied practically.Based on the sparseness of the speech signal,the sparsity method is proposed to determine the optimal order and compared with the weighted variance method.The results show that the sparsity method of determining the optimal order in this paper is faster.The results are more accurate and effective.(2)Multilevel FrFT speech enhancement algorithm research.After studying the problem that traditional transform domain methods cannot completely separate speech and noise in non-stationary environments,a multi-stage FrFT speech enhancement algorithm based on sparse is proposed in this paper.First,the sparsity method is used to calculate the optimal transform order of each stage,and a multi-stage fractional Fourier transform is performed on the noisy speech signal.Then the Minimum Mean Square Error method is used to calculate the filter frequency response of each fractional order domain,and iterative optimization is performed to determine the optimal filter frequency response of the fractional domain with a threshold.Finally,according to the determined frequency filter response of the optimal fractional domain,noisy speech signals are processed in different fractional order domains to obtain enhanced speech.Through experimental verification,the algorithm in this paper can effectively remove noise,reduce distortion,and improve the Signal-to-Noise Ratio.(3)In this paper,the objective evaluation method,such as Signal-to-Noise Ratio,LSD and PESQ,are used to evaluate the speech enhancement quality.Then the algorithm in this paper is compared with several traditional algorithms.The experimental results show that the multi-stage FrFT speech enhancement algorithm based on sparsity in this paper has a good noise reduction effect,can greatly improve the Signal-to-Noise Ratio,and is superior to several other algorithms.