Rate distortion theory promises that autoregressive sources can be encoded optimally at small distortions (high rates) by a source coder with infinite encoding delay and zero delay at the decoder. However, for instrumentable systems with finite encoding delay and an unmatched code generator or for operation at low rates, decoding delay may provide a performance increment. The alphabet constrained approach to data compression allows delay at both the encoder and the decoder, and Sethia and Anderson incorporate delay in a tree coder code generator by combining a weighted linear interpolation scheme with DPCM. This system, called interpolative DPCM (IDPCM), was shown to outperform DPCM at rate 1 b/sample for several synthetic source models. In the present work, we use minimum mean squared error (MMSE) fixed-lag smoothing in conjunction with DPCM to develop a code generator employing delayed decoding. This smoothed DPCM (SDPCM) code generator is compared to DPCM and IDPCM code generators at rates 1 and 2 b/sample for tree coding several synthetic sources and to a DPCM code generator at rate 2 b/sample for speech sources. The (M,L) algorithm, which retains only the M best paths to depth L, is used for tree searching, and SDPCM outperforms IDPCM and DPCM at rate 2 b/sample for the synthetic sources with M = 1, 4, 8, and 12, and at rate 1 b/sample with M > 4. For speech, SDPCM provides a slight improvement in MSE over DPCM codes that is also evident in sound spectrograms and informal subjective listening tests. The models upon which the fixed-lag smoother is based must be chosen appropriately to achieve good SDPCM performance.