The Rearch On The Technology Of Melpe Speech Coder

Speech coding is one of the important supporting technologies of the modern speech comunication system, and the research of very low speech coding algorithm is an important research direction of speech coding area. This paper concerns on the key technologies of the MELPe speech coding algorithm. In this paper,I try to complete the job below:(1) Study and realize the algorithm of speech enhancement. This method transforms the noise speech signal to audio frequency domain, then according to Bark critical bandwith, it divides the signals into a group of frequency segments which do not lap over each other. In every frequency segment, it gets the increment based on the Signal to Noise ratio, and according to the different characteristics of that noise and speech are distributing differently under dissimilar frequency power, it makes the amplitude spectrum of the noisy speech multiply with increment to restrain the noise and to improve the subjective sense quality of the speech.(2) Study the estimating algorithm method of speech parameters. These parameters include pitch period,gain,voice/unvoice decision,Fourier amplitude,linear spectrum pair, etc. In the algorithm of 1200bps and 600bps, my paper focus on the research of the smooth algorithm of the pitch period.(3) The important part of my research is the parameter quantization algorithm. In 600bps algorithm, in order to further lower Super Frame bit, we adopt classified and patterned parameter quantitative scheme. This scheme not only sorts Super Frame, but also classifies patterns according to the differences of adjacent frames. In same pattern but different mode or same category different mode, we adopt different quantitative scheme. In mult-frame pitch period parameter quantifying, we adopt quantitative scheme according to fundamental frequnecy track, so as to use less bits to further quantify fundamental frequency.(4) We put forward 1200bps and 2400bps speed conversion algorithm. By setting buffer length, we can implement the convention of two speeds.(5) In this paper, a new mixed excitation generation method is studied. This method improves MELP algorithm's excitation generation method and adopts changing mixed verge method according to band pass voiced intensity.(6) We achieve 2400bps/1200bps/600bps code and decode algorithm. The experiment result shows that three algorithms all have higher clarity and clearness. 2400bps algorithm bears higher naturalness.