Research Of Low-bit-rate Speech Coding Algorithms Based On Melp

Due to the limited channel resources, as well as the needs of military communications, high-quality low-bit-rate speech coding technology has a great potential of development. Both the rate of 2.4kbps and the lower rate of speech coding algorithms have become very hot in the field of research recently.As the new speech coding standard at 2.4kbps adopted by the U.S. Department of Defense, Mixed Excitation Linear Prediction (MELP) coder is not only one of the best coders currently, but one of the most potential algorithms. This paper studied the way that the MELP model works. After the analysis of a traditional speech coder at 2.4kbps named LPC-10, it came to the conclusion that the MELP coder employs such superior processing methods as mixed-excitation and vector quantization in comparison with LPC-10. The MELP coder also adopts a speech model which matches the mechanism of human voice more exactly to achieve the improvement of the synthetic speech quality, with the application of new technologies such as self-adapted amplitude-enhancing, etc.Based on the deep research of MELP, this paper came up with a new integer pitch-detecting algorithm which can implement more precise detection with much more simpler computational complexity. This paper also presents an improved MELP model with the bit rate of 1.7kbps, while the synthetic speech quality is clear according to the experiment results on PC-platform with MATLAB.