Characteristic　waveform　(CW)　decomposition　based　on　discrete　cosine　transform　(DCT)　is　proposed　firstly　in　waveform　interpolation　(WI)　speech　coding.　The　complicated　CW　alignment　operation　is　cancelled　by　using　this　method.　Secondly,　a　new　unvoiced/voiced　phase　decision　method　and　the　classification　method　of　voiced　phase　are　proposed　with　respect　to　phase　reconstruction.　Finally,　DCT-WI　coders　at　2.0kbps　and　1.6kbps　are　designed,　respectively.　The　subjective　MOS　test　results　indicate　that　the　2.0kbps　DCT-WI　coder　outperforms　2.4kbps　MELP　coder　and　the　1.6kbps　DCT-WI　coder　also　has　a　good　performance.