Based Eemd Car Voice To Promote Research

In-car human-machine speech-based interactions are becoming increasinglyimportant, and finding more and more applications. However, the performance ofin-car human-machine dialogue is considerably deteriorated by background noisesand other disturbances. Therefore, to develop effective speech front-end processingtechniques used for suppressing the background noises are the key step in order toimprove the speech intelligibility at the in-car noisy environment.To address this problem, authors introduce an in-car speech enhancement (ICSE)method based on EEMD in this paper. Due to the adaptive band-pass filter of constantratio, as well as the different characteristics of speech and noise in IMFs, using thenonlinear least-square estimation and the signal-to-noise ratio (SNR), the optimalweighting coefficients of those IMFs by which the speech signal is dominated can befound out. Then, the enhanced speech signals which in-car noises have been removedare obtained by its reconstruction based on the weighted IMFs.The experiments show that the ICSE method is feasible and effective. The voicequality has been greatly improved. Whether high or low SNR, the waveform and theaudible have shown good results, and the general environment of the car noise hasuniversal applicability. The spectral subtraction and band-pass filter are chosen as thecomparative method. It is evident that the ICSE method is a very effective technologyfor separating clean speech from the in-car noises, and the weighting coefficientsintroduced in this study are efficient for noise reduction as considering the similarityof the signal waveform. It provides a viable solution for in-car speech enhancement.