A mixed harmonic excitation linear predictive speech coding for low bit rate applications

The mixed harmonic excitation linear predictive speech coder (MHE-LPC) is based on a technique derived from the multi-band excitation (MBE) and multi-band linear predictive coding (MB-LPC) types of speech coding algorithms. The MHE-LPC coder has the potential of producing high quality speech at 4.8 kb/s and below. This coder employs a new pitch estimation and a mixed voicing technique that utilizes a number of bands within the 4 kHz speech bandwidth (6-8 bands), each band having a voicing probability that defines voiced and unvoiced portions of the excitation spectrum. In addition to these, a new DCT based LPC and residual amplitude quantization technique has been developed and is described in this paper. The 3.6 kb/s MHE-LPC coder with 14th order LPC filter was found to produce better speech quality than various other low rate speech coding standards. During an informal listening test, the 3.6 kb/s MHE-LPC vocoder produced better performance than the 3.6 kb/s INMARSAT Mini-M coder. The MHE-LPC algorithm can be easily extended to bit rates between 1.2 and 8 kb/s depending on the applications.