Efficient coding of LPC parameters using the ordering property of line spectrum pairs

Abstract

Efficient quantization methods of the line spectrum pairs (LSP) which have good performance and low complexity and memory are proposed. The ordering property of the LSP parameters is utilized in a scalar quantizer and a vector-scalar quantizer of the LSP parameters. The proposed scalar quantization algorithm needs 31 bits/frame which is 3 bits less than in the conventional scalar quantization method with interframe prediction in order to maintain the transparent quality of speech so that the average spectral distortion is less than 1 dB. The improved algorithm is implemented in the QCELP vocoder which is used in a CDMA digital cellular system and its performance is also evaluated. The quantization method using the ordering property of the LSP parameters is applied in the vector-scalar quantization scheme. The improved vector-scalar quantizer achieves an average spectral distortion of 1 dB using 26 bits/frame. The quantization performance has also been shown to be robust across databases and different speakers.