A　novel　characteristic　waveform　(CW)　decomposition　method　based　on　bi-orthogonal　lifting　wavelet　transform　(BLWT)　is　proposed　for　CWI　speech　codec　in　this　paper.　Firstly,　the　complicated　CW　alignment　operations　are　not　necessary　by　using　this　method,　and　as　a　result　the　computational　complexity　can　be　greatly　reduced.　Secondly,　the　additional　delay　of　CW　decomposition　based　on　traditional　wavelet　transform　is　cancelled　by　using　boundary　treatment.　Thirdly,　by　omitting　filter　operations　and　adopting　the　situ　calculation　technique,　the　memory　consumption　of　the　proposed　algorithm　will　be　much　smaller　than　the　traditional　one.　In　addition,　the　multi-resolution　analysis　capability　of　wavelet　transform　is　preserved,　which　will　make　the　parameter　quantization　more　flexible　and　the　scalable　coding　much　easier　to　be　realized.　The　performance　evaluation　shows　that　the　proposed　decomposition　algorithm　could　fulfill　the　requirements　of　high　quality,　multi-rate　speech　compression,　and　is　suitable　for　the　real-time　communication　systems.　The　results　of　MOS　and　A/B　listening　test　show　that　the　performance　of　1.84kb/s　CWI　coder　based　on　this　CW　decomposition　is　very　close　to　that　of　2.4kbit/s　MELP　coder.