In-car speech enhancement based on ensemble empirical mode decomposition

摘要

The performance of the human-machine dialogue in a vehicle environment is greatly deteriorated by background noise. In this paper, the authors present an in-car speech enhancement method (ICSE), which can be employed to effectively improve quality of the in-car speech signals. This speech enhancement method was developed by means of the Ensemble Empirical Mode Decomposition (EEMD). Based on EEMD method, noise-contaminated speech signals can be decomposed into a set of intrinsic mode functions (IMFs). Next, nonlinear least squares estimation and signal-to-noise ratio (SNR) testing are employed to obtain optimal weighting coefficients of the dominant IMFs. Finally, by combining the weighted IMFs, the noise effect can be suppressed sufficiently and the in-car speech signals enhanced significantly. Experimental and simulative results show that the obtained weighting coefficients of IMFs are reasonable and the EEMD technology is effective for extracting clean speech from the noise-contaminated in-car speech.