Glottal codebook constrained iterative Wiener filtering speech enhancement

After deeply researching into the All-Pole based iterative Wiener filtering speech enhancement algorithm, we obtained the result that low quality of the enhanced speech was mainly due to glottal wave distortion, so we probed into the crude way to improve enhanced speech quality by glottal wave restoration. We proposed the glottal codebook constrained iterative Wiener filtering speech enhancement algorithm. This method adapts the clean training speech to get it's glottal wave, and analyses the clean training glottal wave frame by frame pitch synchronously based on LF model and modified turbulent noise model, and then clusters training glottal wave frames according to extracted glottal features using k-means algorithm to produces the glottal codebook, which represents the effective exciting model of clean human speech. We imposes constraints on the glottal wave of enhanced speech during iterative Wiener filtering according as the glottal codebook to ensure glottal wave laid in the clean exciting glottal space. As shown in the simulation experiments, this method can reduce the glottal wave distortion and improve enhanced the speech quality effectively.