| 摘要 |
A method of speech signal processing is disclosed which does not require excessive computation and which operates at low bit rates. The method comprises filtering blocks of digital samples of speech signals by a linear-prediction inverse filter followed by a shaping filter, the coefficients of which are chosen from a codebook of quantized filter coefficient vectors, to obtain a residual signal subdivided into vectors. Each vector is classified by an index q depending on a zero-crossing frequency and a r.m.s. value. Each vector is then normalized first on the basis of a quantized r.m.s. value, and then on the basis of a vector of quantized short-term mean values;the mean-square error occuring in quantizing these vectors with vectors contained in a codebook and forming excitation waveforms is computed. In this codebook the search is limited to a subset of vectors determined by index q and p of each short-term mean vector. The coding signal consists of the index of the filter coefficient vector, of indices q, p, of the quantization index m of the r.m.s. value, and of the index of the vector of the excitation waveforms found to generate the minimum weighted mean-square error. |
