Research And Implementation On Low Bit Rate Speech Coding In Narrowband Communication

Language is the most basic and important mean for people to communicate with others, but because of the redundancy in language, the transmission and storage of speech information in communication system faces a lot of problems. The technologies about low bit rate speech coding, even very low bit rate speech coding which aim at solving the contradictions between bandwidth, the efficiency and reliability of speech transmission have been a hot research topic for its deeply practical significance.Dividing the speech signal into voice and unvoice, and choosing different excitation sources to synthetic speech can simulate the periodicity of the original speech signal effectively; Dividing the frequency band of the speech signal into several sub-bands and calculating the strength of voice in every sub-band can describe the information about voice and unvoice more accurately; using sinusoidal excitation model as the excitation source of voice speech can describe the amplitude characteristics of speech as well as the periodicity, and can make the synthetic speech more clear. The Sinusoidal Excitation Linear Prediction speech coding algorithm raised by Tsinghua university uses all the three technologies above, which can simulate natural language very well in low speech rate and the quality of the synthetic speech is also high. So the algorithm of SELP is one of the algorithms which have great potential.Firstly, based on the deeply study of the technologies in the algorithm of SELP, this thesis detailed explains the calculated and quantitative methods of each parameter in the coding process of SELP algorithm, and describe the process of Speech synthesis in detail. Then designed and implemented the1.2Kb/s SELP algorithm using the technologies of increasing the long of every frame, structuring super frame and parameters interpolating. Based on MELP open source code, using C language to realize the2.4Kb/s and1.2Kb/s SELP algorithm in PC, comparing the synthetic speech and the original speech in time domain and frequency domain and evaluate the quality of the synthetic speech using the method of Mean Opinion Score, the result can meet the basic requirements of communication system.