| Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 90 === Many speech coding standards are based upon code-excited linear prediction (CELP), and it is desirable to enhance signal performance by using layered coding methods that are compatible with this base coder.
In this thesis, a hybrid bit rate scalable coding method is proposed for MPEG-4 CELP that addresses similar functionalities of MPEG-4 CELP Bit Rate Scalability, and is also layered structured to enhance signal performance. It offers two modes of coding in different situations: differential position waveform-based coding (DPWBC) for high coding bit rate of the base layer and partial decoding of MPEG-4 CELP Bit rate Scalability (PDMCBRS) for low coding bit rate of the base layer. Instead of using analysis-by-synthesis coding method, the DPWBC method codes the waveform of the residual signal. In the residual signal, continuous positive (or negative) samples are regards as one signal to be coded. Each signal is coded by using its position, sign, width and magnitude. The coding order of signals are arranged according to signal’s group and position to reduce coding bit rate and increase coding accuracy. The PDMCBRS method employs partial decoding of the MPEG-4 CELP Bit Rate Scalability to gradually enhance speech performance. Either modes of the proposed coding method has smaller bit rate step than MCBRS, so the concept of successive refinement is more closely approached.
Experiments show that DPWBC enhances the performance more effectively than MCBRS at high bit rate and noisy background; and PDMCBRS has the same performance as the MPEG-4 CELP Bit rate Scalability. So the proposed coding method employing DPWBC and PDMCBRS performs better than MPEG-4 CELP Bit rate Scalability and has smaller bit rate step of enhancement.
|
|---|
