Enhanced modeling of discrete spectral amplitudes
Keith A. Teague, Walter D. Andrews, Bradley E. Walls · 2002
Speech coders such as those based on multiband excitation (MBE) and STC models depend on having available an accurate representation of a discrete set of significant spectral amplitudes. Normal linear predictive (LP) spectrum representations model adequately only the general spectral envelope, not individual spectral amplitudes. A technique for enhancing the accuracy of a LP spectrum representation through the use of cubic spline interpolation and amplitude compression is described. The technique is applied to a 2400 bps MBE-type coder (EMBE 2400) and shown to result in significantly lower spectral distortion than a simple LP model alone.