(A) study on a multi-rate speech coding algorithm

Abstract

In this thesis, we study a multirate vocoding algorithm that is compact and yields good quality of speech at bit rates of 2.4, 4.8 and 9.6 kbits/s. The resulting multirate algorithm utilizes the conventional LPC algorithm with the adaptive pre-/de-emphasizer based on V/UV decision as a 2.4 kbits/s coder, and the improved RELP algorithm as 4.8 and 9.6 kbits/s coders. First, we suggest the adaptive pre-/de-emphasis scheme based on V/UV decision. This scheme is incorporated in the LPC algorithm. The resulting synthetic speech with the scheme is more intelligible and natural than that without it. Next, we propose a new coding scheme that incorporates the values of minimum and maximum step sizes obtained by the average residual error energy (ERA) of each frame. The new coding scheme increase the dynamic range and results in increased SNR. Also, three full-band residual reconstruction schemes are proposed to improve the speech quality of the RELP vocoder. These include the multiband spectral folding scheme, the scheme of using both the spectrally folded signal and LPC excitation signal, and the scheme of using the multiband spectrally folded signal and LPC excitation signal. It has been found that among the three schemes, the last one has the best performance. It produces no roughness and little tonal noise. Finally, the hardware implementation of the resulting multirate algorithm using the TMS320 digital signal processor family is considered.