Object-based　audio　techniques　have　become　common　since　they　provide　the　flexibility　for　personalized　rendering.　In　this　paper　a　multi-stage　encoding　scheme　for　multiple　audio　objects　is　proposed.　The　scheme　is　based　on　intra-object　sparsity.　In　the　encoding　phase　the　dominant　Time　Frequency　(TF)　instants　of　all　active　object　signals　are　extracted　and　divided　into　several　stages　to　form　the　multi-stage　observation　signals　for　transmission.　In　the　decoding　phase　the　preserved　TF　instants　are　recovered　via　Compressed　Sensing　(CS)　technique,　and　further　used　for　reconstructing　the　audio　objects.　The　evaluations　validated　that　the　proposed　encoding　scheme　can　achieve　scalable　transmission　while　maintaining　perceptual　quality　of　each　audio　object.