In this paper block constrained trellis coded vector quantization (BC-TCVQ) is presented for quantizing the line spectrum frequency parameters of the wideband speech codec. Both a predictive structure and a safety-net concept are combined into BC-TCVQ to develop the predictive BC-TCVQ The performance of this quantization is compared with that of the linear predictive coding vector quantizer used in the AMR-WB codec, demonstrating reductions in spectral distortion.