EUROSPEECH 2001 Scandinavia
This paper describes a hybrid sub-band speech coding scheme based on sinusoidal coding and CELP. Purely voiced speech is encoded using sinusoidal coding techniques and phase information is selectively transmitted. For mixed and unvoiced speech, the lower band is processed by sinusoidal coding algorithms while the upper band is encoded using CELP. To accommodate the extra bandwidth required by the encoded CELP parameters, the phase information is disregarded. The proposed coder is enhanced by sub-band discrete all-pole modeling and a voicing detection technique based on an analysis-by-synthesis approach. An efficient adaptive spectral shaping technique based on bandwidth widening in the LSP domain is employed. The proposed technique is capable of producing high quality speech at 4.1 kbit/s.
Bibliographic reference. Ho, M. S. / Molyneux, D. J. / Cheetham, B. M. G. (2001): "A hybrid sub-band sinusoidal coding scheme", In EUROSPEECH-2001, 1985-1988.