Composite Permutation Coding with Simple Indexing for Speech/Audio Codecs
Shinya Abe, Kei Kikuiri, Nobuhiko Naka · 2007
This paper proposes a vector quantization (VQ) method based on composite permutation coding for transform audio coding. VQ is widely used for audio data compression. It requires mean square error computation or a similar metric for finding the nearest neighbor in the codebook, which generally incurs a lot of operations. To reduce such operations, we focus on the permutation representation and easy indexing of vectors in the codebook. The proposal consists of constrained composite permutation codes, which are distinguished by the number of components quantized into each quantization level. This scheme makes the output bit stream take the same form as a parallel array of scalar quantization (SQ). Simulation results show that the proposal almost matches the performance of VQ at 2 bit/scalar bit-rates with lower computational complexity. Its structure yields the efficient representation of tones that are important for auditory perception.