Improved Codebook-Based Speech Enhancement Based on MBE Model

This paper provides an improved codebook-based speech enhancement method using multi-band excitation (MBE) model. It aims to remove the noise between the harmonics, which may exist in codebook-based enhanced speech. In general, the proposed system is based on analysis-withsynthesis (AwS) framework. During the analysis stage, acoustic features are extracted including pitch, harmonic magnitude and voicing from noisy speech. These parameters are obtained on the basis of the spectral magnitudes obtained by codebook-based method. During the synthesis stage, different synthesis strategies for voiced and unvoiced speech are employed. Besides, this paper introduces speech presence probability to modify the codebook-based Wiener filter so that more accurate acoustic parameters can be obtained. The proposed system can eliminate noise not only between the harmonics, but also in the silent segments, especially in low SNR noise environment. Experiments show that, the performance of the proposed method is better than traditional codebook-based method for different types of noise.