Complexity reduction in Karhunen-Loeve transform based speech coder for voice transmission

For gain-shape split vector quantization, an incremental M (IM) search algorithm is proposed. In the conventional M search, every possible combination of subvectors is evaluated. However, in the proposed IM search, an intermediate subvector is constructed by merging adjacent subvectors in an incremental and recursive way. Based on the IM search, the high-dimensional Karhunen-Loeve Transform (KLT) based speech coder is designed in practice. IM search requires 1800 times lower complexity than M search in designing a high-dimensional KLT-CVQ. Based on the proposed KLT based speech coder, a novel voice over IP (VoIP) system is designed, which produces superior voice quality and rate-distortion performance to conventional VoIP systems.