One Low-rate Speech Coding Algorighm Based On Linear Prediction And Its Simulation

With the rapid development of Information society and Communications technology, frequency resources needed are becoming more and more precious. Over the years, it has been the goal that compressing the transmission bandwidth of speech signal or reducing the transmission rate of telephone channel. In achieving this target, speech compression coding technology plays an important role. At present, Low Bit Rate Speech Coding which we will research is especially of important significance at the situation where the transmission rate is limited very strictly.Among many studies of 1ow-rate speech coding algorithms, MELP (Mixed Excitation Linear Prediction) algorithm is still based on the traditional LPC (Linear Prediction coder) parametric model, but It's speech synthesis model is rather close to true pronunciation mechanism, integrating the elite of several speech coding technology, such as Linear Prediction, Multi-band excitation, Mixed Excitation, Prototype Waveform Interpolation and Vector Quantization. So its speech quality is improved compared to LPC coder. MELP algorithm is an ideal low-rating voice coding scheme relatively, also it is the emphasis that this paper will research. After researching, we found that MELP algorithm is excellent in deed, and it is a new 1ow-rate speech coding algorithms, and it has a great value in the future actual application.This paper at first surveys the principle of the speech compression coding and the development and the current progress of the speech coding technology, and the new five key technologies and the principle of the traditional MELP vocoder. Meanwhile, the procedure of coding and decoding is further researched. The detailed designing procedures of MELP are introduced, in that time. We have improved the algorithm, we also have done the work of simulating analysis and researched the method of completing this Arithmetic of MATLAB programming language. Comparing the speech qualities of the coded and un-coded,this paper gives out the results of the experiment,presents the conclusion and points out the next research direction. The conclusion shows that show that the speech quality of the improved 2.4kb/s coding algorithm achieves high intelligibility, clearness, and relative naturalness.Based on the traditional MELP algorithm, we can study it from the following three aspects.Firstly, we can optimize the algorithm, and change root operations into square operations; Secondly, we don't transmit Fourier series amplitude values, namely cancelling the calculation and quantification of Fourier series amplitude; Thirdly, by Sliding Window, we can improve the computational model of the pitch.In short, the first two methods are simplified to MELP algorithm, they can accelerate the speech of computer operation procedure, reduce the complexity of the algorithm, the third method makes the improved speech in a certain degree eliminate the noise which results from the inaccurate voiced and unvoiced speech, and makes the pitch more accurate.