Title :
Efficient vector quantisation of LPC parameters for noisy channels
Author :
Skinnemoen, Harald ; Perkis, Andrew
Author_Institution :
Dept. of Telecommun., Norwegian Inst. of Technol., Trondheim, Norway
Abstract :
Line spectral pairs (LSP) are, because of their advantageous properties, commonly used to represent the spectral information in a linear predictive (LP) speech coder. However, the quantised spectrum is relatively sensitive to channel errors. This paper describes a robust quantiser for LSPs transmitted over noisy waveform channels, based on an optimisation of the source coder over the modulation scheme. Modulation organized vector quantisation, MOR-VQ, jointly optimises a codebook and its index assignment, governed by the defined modulation symbol constellation. The system closely follows the optimum performance theoretically available (OPTA). Moreover, the MOR-VQ scheme allows for efficient codebook search strategies
Keywords :
learning (artificial intelligence); linear predictive coding; modulation; noise; optimisation; self-organising feature maps; source coding; spectral analysis; speech coding; telecommunication channels; vector quantisation; Kohonen learning; LPC parameters; MOR-VQ; channel errors; codebook search; line spectral pairs; linear predictive speech coder; modulation organized vector quantisation; modulation symbol constellation; noisy waveform channels; optimisation; optimum performance theoretically available; quantised spectrum; robust quantiser; self organised map; source coder; spectral information; vector quantisation; Amplitude modulation; Channel coding; Forward error correction; Linear predictive coding; Modulation coding; Oral communication; Process design; Quadrature amplitude modulation; Speech; Vector quantization;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 1994. ICASSP-94., 1994 IEEE International Conference on
Conference_Location :
Adelaide, SA
Print_ISBN :
0-7803-1775-0
DOI :
10.1109/ICASSP.1994.389247