Research Of Speech Codec Based On CELP And It's Implementation On DSP

In the information age, how to obtain high quality and correct information timely becomes the focus of research. Speech communication is one essential and effective way to obtain information. And the key technology of speech communication is compressing coding, i.e.. how to obtain more information by less bits. The research of speech codec and the establishment of standard will have great influence upon the domain of communication and IT.Wave coding and parameter coding are used conventionally but they have disadvantages either high bit rate or bad quality of synthesis speech. Syncretize method of coding based on CELP (Code Excited Linear Prediction) is researched in the paper. It can conquer the disadvantages above. The CELP algorithm includes many important module such as LP(Linear Predictive) analysis, adaptive codebook, random codebook and gain quantization of code etc. The main difficulties of real time implementing are complexity of the algorithm and store. We have improved the algorithm of LP, and written a class of LP by C++. It can reduce instructions and make it more efficient. Some CELP modules are improved such as the search of stochastic codebook so the complexity is reduced. We have a deep study of RIFF file format and put forward its operation. So we can have simulation. The time-domain and frequency-domain of speech can be analyses. Thus the quality of speech can be compared.Because DSP have many advantages such as small size, high speed, especially adapt to arithmetic. So we design software and hardware of real-time system based on TMS320C5410 of TI Company. Main algorithm and peripherals are designed. A foundation to realize a speech codec and embeded-system is set up.At last, evaluation of speech coding in our design is discussed.