On Soft-Decision Demodulation for PCM- and DPCM- Encoded Speech