This　paper　proposes　a　new　approach　for　generating　smart　3D　speech　animation.　The　basic　idea　is　to　synthesize　the　animated　faces　using　prosodic　information　edited　by　user　with　a　kind　of　markup　language.　The　proposed　technique　takes　advantage　of　both　performance-driven　and　parameter-driven　approaches.　So　it　greatly　reduces　the　workload　of　manual　modeling　used　in　the　traditional　key　frame　animation　and　the　animation　generating　process　can　be　easily　control.　To　relate　the　prosody　text　with　the　3D　animation,　our　technique　builds　up　a　parametric　model　based　on　the　exponential　formula.　It　takes　the　pre-obtained　3D　dynamic　visemes　and　prosodic　tag　recorded　in　CPML　(Chinese　Prosody　Markup　Language)　as　input　data,　and　outputs　a　segment　of　vivid　speech　animation.　Experimental　results　show　that　(1)　the　proposed　technique　synthesizes　animation　of　different　effects　depending　on　the　availability　with　the　prosodic　information,　and　(2)　the　new　technique　produces　realistic　results　using　less　data　than　the　conventional　methods.　©　2008　IEEE.