Coding of unquantized spectrum sub-bands in superwideband audio codecs

We present several new methods for coding of spectrum coefficients in embedded audio codecs. In audio codecs, the spectrum is often divided in sub-bands, and quantized using a vector quantization. In many applications, the limited bit-budget allocated to the vector quantization is not sufficient to cover all the sub-bands, and the distortion of the spectrum in the unquantized sub-bands frequently causes audio quality degradation. The presented methods have been designed to improve the representation efficiency of the spectrum in these sub-bands. They consist of a preprocessing of the spectrum before the vector quantization, a correction of the encoded spectral envelope after the quantization, and a filling of the unquantized spectrum sub-bands. These methods have been used in the superwideband embedded extensions of the audio codecs G.711.1 Annex D and G.722 Annex B, recently standardized by the ITU-T.