In-car Speech Enhancement Using Ensemble Empirical Mode Decomposition

The performance of the human-machine dialogue at in-car environment is considerably deteriorated by background noises and other disturbances. In this paper, the authors present an in-car speech enhancement (ICSE) method to improve quality of speech signals suffering the in-car noises. The method is based on a novel signal processing technology called the ensemble empirical mode decomposition (EEMD). By using EEMD, the noisy speech signals are decomposed as a set of intrinsic mode functions (IMF). Then, the nonlinear least-square estimation and signal-to-noise ratio (SNR) are employed to find out the optimal weighting coefficients of those IMFs dominating the speech signals. Finally, the enhanced speech signals, where in-car noises are suppressed, are obtained by the reconstruction technique based on the weighting IMFs. Results of the work show that, EEMD is an effective technology of separating pure speech from in-car noises, and the weighting coefficients proposed in this study are also effective for noises reduction as considering the similarity of the signal waveform.