The　principal　idea　of　this　research　for　visual　speech　synthesis　realism　is　that　oral　features　localization　provides　the　precise　geometrical　information　for　oral　images　of　speech　sound.　In　this　research,　Hierarchical　Active　Shape　Model　(HASM)　is　used　as　local　texture　model.　In　the　structuring　local　texture　model,　the　local　texture　model　is　the　main　clue.　The　optimal　oral　features　localizations　are　decided　by　Mahalanobis　distance.　This　research　utilizes　variable　step　strategy,　variable　angle　strategy　and　oral　images　clustering　strategy　to　greatly　improve　the　accuracy　and　efficiency　of　inter　lip　localization　and　special　lip　shape.　The　result　shows　that　the　accuracy　and　efficiency　are　up　to　90%.