A New Optimization Engine for the LSF Vector Quantization
Jian Guo Tang, Jingsong He, Lu Huang · 2007
The speech signal line spectral frequencies (LSF) vector quantization is one of the most important issues in speech coding systems, while the codebook design is the key problem, which impacts the synthesis speech quality severely. A new optimization mechanism named vector space scale-stretching (VSSS) was presented in this paper. By introducing nearest neighbor partition (NNP) control factor based on the scale-stretching technique, the codebook centroid track was optimized in a comparatively low dimension space. The new optimization mechanism reduced the search space of evolutionary computation and made the evolutionary operation more maneuverable. The experiments show that the VSSS is effective, and the new codebook design algorithms are robust.