Speech Enhancement Research Based On Improved Phase Spectrum Information And Phase Reconstruction

In the past three decades,speech enhancement has been an active research field,which can extracts the“pure speech”from the noisy speech affected by various kinds of noise to improve the perceived quality and intelligibility of speech.Speech enhancement can be applied in areas such as voice communication,human-computer interaction,and artificial intelligence.At present,researchers have proposed various speech enhancement algorithms based on spectral amplitude information.However,these algorithms introduce some speech distortion while suppressing background noise,thus reducing the speech intelligibility.Recent studies have shown that phase information has a significant impact on improving the intelligibility of speech signals.Based on the introduction of the speech enhancement algorithms with phase spectrum compensation and phase reconstruction,a speech enhancement algorithm with improved phase spectrum information and phase reconstruction is proposed.The main work of this paper is as follows:1.Briefly introduce the purpose,significance and development of speechenhancement,expounds the principle of traditional speech enhancement algorithm,analysis the noise estimation method,fundamental frequency estimation and related experimental simulation.2.Aiming to the problem that the compensation factor in the traditional phase spectrum compensation algorithm is fixed and the phase spectrum of the noisy speech can not be flexibly compensated,a speech enhancement algorithm with improved phase spectrum information is proposed.Firstly,the Sigmoid phase-spectral compensation function which is based on the signal to noise ratio of each frame input speech is presented in this paper,it can flexibly compensate the phase spectrum of the noisy according to the change of noise;next,Estimate noise power spectrum through combine a priori SNR estimation of the improved DD and speech presence probability algorithm(SPP).Finally,Wiener filtering is applied to improve speech enhancement effect by combining the new speech presence probability noise power spectrum estimation and phase spectrum compensation.Compared to the traditional phase spectrum compensation(PSC)algorithm,the improved algorithm can effectively suppress the noise in the audio signal,and then enhance the perceived quality of the speech and improve speech intelligibility.Firstly,the Sigmoid phase-spectral compensation function which is based on the signal to noise ratio of each frame input speech is presented in this paper,it can flexibly compensate the phase spectrum of the noisy according to the change of noise;next,Estimate noise power spectrum through combine a priori SNR estimation of the improved DD and speechpresence probability algorithm(SPP).Finally,Wiener filtering is applied to improve speech enhancement effect by combining the new speech presence probability noise power spectrum estimation and phase spectrum compensation.Compared to the traditional phase spectrum compensation(PSC)algorithm,the improved algorithm can effectively suppress the noise in the audio signal,and then enhance the perceived quality of the speech and improve speech intelligibility.3.Aiming to the problem that the traditional phase reconstruction algorithm uses the same method to reconstruct the phase of the unvoiced and voiced segments of the speech signal,which makes the effect of phase estimation not obvious,a new method for improving the phase reconstruction by using the signal-to-noise ratio information and the time-frequency feature is proposed.Firstly,the time-frequency characteristics related to phase distortion are introduced and the decision threshold is calculated.Then,the signal-to-noise ratio information is used to calculate the phase deviation between the noisy speech and the pure speech.The two comparisons further estimate the speech phase of the unvoiced segment and the voiced segment,which can effectively improve the speech intelligibility.Finally,the reconstructed phase is combined with the amplitude estimation of the improved binary hypothesis model and speech enhancement is carried out.Experiments on different noisy speech signals under different noise conditions show that the signal-to-noise ratio of the new algorithm is significantly improved,and the speech perception evaluationindex is better than the traditional algorithm,which shows that the speech distortion is reduced and the speech intelligibility is improved.