Variable rate speech compression based on exact modeling and waveform vector quantization