Low-rate Speech Coding Algorithm Based On Linear Predictive Coding

In modern society, the communication bandwidth is more and more precious. How to reduce the transfer bandwidth of voice signals is always what we want and the voice coding technology is very important to achieve this goal. With the increasing of network users, the integrating and diversifying of network service the contradiction among bandwidth, system capacity and service quality are more and more obvious. traditional speech coders can not satisfy the requirement of transmit bandwidth which is becoming more and more congested. So, how to reduce the bit-rate of speech without notably degrading its perceptual quality is a question placed at the front of the researchers. During the last ten years, a lot of high-quality middle bit-rate coders (4.8kbps to 16kbps) had been successfully used in various digital services. Now the low bit-rate coders (less than 2.4kbps) and ultra low bit-rate coders (less than 1.2kbps) become the focuses in many researches. First, we introduce the actuality of speech coders in chapter 1 and give the principle of linear prediction analysis in chapter 2. Then we analyze the shortcomings of conventional LP analysis and introduce a novel LP model in chapter 3. Simulations show that the new LP model can significantly improve the spectral match than conventional LP model. In chapter 4 we propose a modified version of NCCF pitch extraction method, which was found to be more accurate than standard MELP pitch extraction method. At last we give a 1.2kbps coder algorithm using the new LP model and modified NCCF pitch extraction method in chapter 5, matlab simulation show that the reproduced speech quality of the new vocoder is near to standard 2.4kbps MELP vocoder.