Weighted trajectory modeling of LSF for low bit rate speech coding
Bin Zhang · Journal of Lanzhou University · 2010
To increase the fitting accuracy of LSFs' trajectories for high-energy frames in low bit rate speech coding,a novel trajectory modeling of LSFs was proposed.This new modeling was based on the trajectory compression of superframe's parameters and used the time weighting based on the importance of frames in a superframe.The modeling calculated the perceptual enhanced trajectories of LSFs by minimizing the time weighting mean square error.Then these feature vectors,obtained by sampling the perceptual enhanced trajectories,were quantified and encoded.Subjective tests show that the new modeling enhanced the MOS of the reconstructed speech by more than 0.02,and effectively improved the fitting accuracy of LSFs' trajectories for perceptual important frames.