A　sinusoidal　audio　analysis　and　synthesis　model　based　on　Empirical　Mode　Decomposition　(EMD)　is　proposed　in　this　paper.　The　EMD　is　an　adaptive　method　for　the　analysis　of　nonstationary　signals　and　it　has　a　unique　filtering　characteristic.　The　Intrinsic　Mode　Functions　(IMFs)　which　derived　from　EMD　fully　retain　the　non-stationarity　of　original　signal.　Because　the　high　frequency　components　of　audio　are　noise-like　and　less　perceptually　relevant,　the　original　audio　is　divided　into　noise　and　harmonics　in　the　proposed　model　by　using　a　combination　of　different　IMFs.　Prior　to　parametric　modeling,　the　audio　is　first　classified　to　reduce　the　affect　of　noise　components　for　sinusoids　extraction,　and　then　the　matching　pursuit　algorithm　combining　with　psychoacoustic　model　and　fine　frequency　searching　method　is　applied　for　extracting　sinusoidal　components.　Finally,　objective　measurements　of　perceived　audio　quality　(PEAQ)　are　performed　to　evaluate　the　effectiveness　of　this　model　for　the　audio　synthesis.