In　recent　years,　sound　source　localization　methods　based　on　Direct　Path　Interchannel　Phase　Difference　(DP-IPD)　estimation　have　gained　considerable　attention,　owing　to　their　outstanding　performance　in　noisy　environments.　However,　existing　methods　face　several　challenges　when　processing　data　from　multi-element　arrays,　as　the　network　must　learn　the　complex　mapping　relationships　between　signals　and　features　across　multiple　microphone　pairs.　These　mapping　relationships　share　a　certain　degree　of　similarity,　which　makes　it　challenging　for　the　network　to　differentiate　them,　ultimately　impacting　localization　accuracy.　Furthermore,　the　time-varying　nature　of　spatial　cues　caused　by　a　moving　sound　source　can　further　degrade　the　performance　of　localization　methods.　To　tackle　these　challenges,　this　paper　introduces　a　Time-Frequency　Feature　Enhanced　Convolutional　Recurrent　Neural　Network.　By　incorporating　a　frequency　attention　convolution　module　and　a　gated　convolution　module,　the　proposed　network　adaptively　handles　the　mapping　of　signals　to　features　across　different　microphone　pairs　while　enhancing　its　ability　to　extract　local　temporal　context.　This　approach　improves　localization　accuracy　for　moving　sound　sources　in　noisy　environments.　Extensive　experimental　results　show　that　the　proposed　method　significantly　outperforms　state-of-the-art　approaches　on　both　simulated　and　real-world　datasets.　©　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.　2025.