Data-driven　computing　and　using　data　for　strategic　advantages　are　exemplified　by　communication　systems,　and　the　speech　intelligibility　in　communication　systems　is　generally　interrupted　by　interfering　noise.　This　interference　comes　from　the　environmental　noise,　so　we　can　reduce　them　intelligibility　by　masking　the　interested　signal　[1,　2].　An　important　work　in　communication　systems　is　to　extract　speech　from　noisy　speech　and　inhibiting　background　noise.　In　this　paper,　the　subspace　algorithm　theory　is　introduced　into　a　speech　noise　reduction　system.　We　first　analyze　the　principle　of　LMS　adaptive　speech　noise　reduction　algorithm　with　the　subspace　algorithm,　and　then,　we　merge　the　subspace　algorithm　into　the　VS-LMS　algorithm　and　propose　a　combined　algorithm　for　an　adaptive　speech　noise　reduction　system.　Furthermore,　we　analyze　the　combined　algorithm,　which　can　decrease　musical　noise,　as　well　as　generate　a　suitable　step-size　factor　to　resolve　the　contradiction.　This　issue　cannot　be　resolved　by　the　current　LMS　algorithm　[31],　which　has　less　convergence　speed　and　larger　residual　noise　than　our　system.　Our　simulation　results　demonstrate　that　our　algorithm　can　get　3　to　10　times　better　than　original　algorithm　in　low　SNR　(-5　similar　to　0db)　and　high　SNR　(0　similar　to　+　5db).