.hybrid-melp/celp Voice Compression Coding Algorithm

Speech compression coding is an important component of signal processing technology, which integrates a lot of ideology, theories and practical methods of signal processing from fundamental area to front line. As a mainstream of signal processing area, it greatly boosts the development of signal processing techniques. Nowadays, speech compression coding algorithm has been divided into two main streams, one stresses coding the parameters in spectrum domain, and the other focuses on coding the waveform in time domain. Main representatives of the above two algorithms are Mixed-Excitation Linear Prediction coding algorithm and Code-Excitation Linear Prediction coding algorithm respectively. The major character of MELP algorithm is that it uses voicing decisions for different sub-bands, and produces LP excitation by mixing voiced ingredients and unvoiced ingredients according to relative voicing strength of every sub-band respectively, among which the voiced excitation is represented by methods of Fourier spectrum magnitudes. On the other side, the character of CELP algorithm is that it does not distinguish voiced or unvoiced speech, and it uniformly uses time domain waveform matching method based on AbS principle to produce LP excitation, which is represented by vector codes. This thesis discusses the principles and concrete implementation methods of both MELP and CELP algorithm. It also simulates the implementation of the above two algorithms based on Matlab and principally, it's Simulink tools. Moreover, a new simplified MELP pitch extraction algorithm is presented in this thesis to solve the problem of the complex mechanism of standard MELP algorithm. The MELP coder is fairly fit for reconstructing a strong periodicity of steady voiced regions. On the other hand, the CELP coder does it well at representing irregular features of weakly voiced and unvoiced speech. The above observations bring to the ideas of combination of the two coder methods, i.e., using CELP, a waveform coder in unvoiced, transitional and...