The LPC-type speech coders require efficient quantization and coding of the LPC spectral parameters in order to operate at low bit-rates. We present here results from an investigation to the Vector Predictive Quantization (VPQ) approach for the compression of the LPC spectral information. Reduction of the computation requirements of the VPQ scheme by means of Split-Vector Quantization is considered. The distortion-rate characteristic of the coding scheme incorporating the Split-Vector Quantizer is estimated, based on log-spectral distortion measures. Our results indicate that this coding scheme can operate at 20 to 22 bits/frame without any perceptually significant distortions in the synthesized speech.