This　paper　proposes　a　multiple　audio　source　separation　method　by　using　the　intra-object-sparsity　(in　each　frame,　the　energy　of　an　audio　signal　concentrates　on　small　number　of　time-frequency　instants)　encoding　framework.　Specifically,　by　applying　the　intra-object-sparsity　of　audio　signal,　each　source　is　encoded　to　obtain　a　sparse　representation　of　it　while　preserves　the　major　energy　of　the　original　signal.　Since,　most　of　the　multiple　source　separation　algorithms　for　speech　sources　can　be　extended　to　the　audio　sources.　The　combination　of　the　intra-object-sparsity　encoding　framework　and　source　separation　method　can　effectively　eliminate　the　cocktail　party　problem　which　lead　to　bad　separation　quality.　The　evaluations　reveal　that　the　proposed　method　achieves　a　higher　separation　quality　compared　with　the　existing　techniques　and　robust　over　different　types　of　audio　signals.　©　2017　IEEE.