{"id":5207,"date":"2026-05-20T16:59:30","date_gmt":"2026-05-20T08:59:30","guid":{"rendered":"https:\/\/www.15zhi.net\/blog\/?p=5207"},"modified":"2026-05-20T16:59:30","modified_gmt":"2026-05-20T08:59:30","slug":"202605-%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-sound-event-detection-and-localization-with-distance-estimation","status":"publish","type":"post","link":"https:\/\/www.15zhi.net\/blog\/202605-%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-sound-event-detection-and-localization-with-distance-estimation\/","title":{"rendered":"202605 \u8bba\u6587\u7814\u8bfb-Sound Event Detection and Localization with Distance Estimation"},"content":{"rendered":"\n

\u4f5c\u8005<\/strong>\uff1aDaniel Aleksander Krause, Archontis Politis, Annamaria Mesaros<\/p>\n\n\n\n

\u65f6\u95f4<\/strong>\uff1a2024<\/p>\n\n\n\n

\u6765\u6e90<\/strong>\uff1a\u00a0IEEE Proceedings<\/p>\n\n\n\n

\u80cc\u666f<\/h2>\n\n\n\n

\u58f0\u97f3\u4e8b\u4ef6\u68c0\u6d4b\u4e0e\u5b9a\u4f4d\uff0cSELD\uff0c\u901a\u5e38\u540c\u65f6\u5b8c\u6210\u4e24\u4ef6\u4e8b\uff1a\u8bc6\u522b\u201c\u53d1\u751f\u4e86\u4ec0\u4e48\u58f0\u97f3\u4e8b\u4ef6\u201d\uff0c\u4ee5\u53ca\u4f30\u8ba1\u8be5\u58f0\u97f3\u4e8b\u4ef6\u7684\u5230\u8fbe\u65b9\u5411 DOA\u3002\u4f20\u7edf SELD \u53ea\u7ed9\u51fa\u65b9\u5411\u4fe1\u606f\uff0c\u4f8b\u5982\u65b9\u4f4d\u89d2\u3001\u4fef\u4ef0\u89d2\u6216\u4e09\u7ef4\u65b9\u5411\u5411\u91cf\uff0c\u4f46\u4e0d\u80fd\u7ed9\u51fa\u58f0\u6e90\u5230\u9ea6\u514b\u98ce\/\u542c\u8005\u7684\u8ddd\u79bb\u3002\u5728\u673a\u5668\u4eba\u542c\u89c9\u3001\u667a\u80fd\u76d1\u63a7\u3001\u589e\u5f3a\u73b0\u5b9e\u3001\u52a9\u542c\u8bbe\u5907\u3001\u4eba\u673a\u4ea4\u4e92\u7b49\u5e94\u7528\u4e2d\uff0c\u4ec5\u77e5\u9053\u65b9\u5411\u5e76\u4e0d\u591f\uff0c\u8fd8\u9700\u8981\u77e5\u9053\u58f0\u6e90\u5728\u4e09\u7ef4\u7a7a\u95f4\u4e2d\u7684\u5b8c\u6574\u4f4d\u7f6e\u3002\u4ee5\u5f80\u58f0\u97f3\u8ddd\u79bb\u4f30\u8ba1\uff0cSDE\uff0c\u7814\u7a76\u591a\u4e0e\u53cc\u8033\u97f3\u9891\u6216\u5355\u72ec\u5b9a\u4f4d\u4efb\u52a1\u7ed3\u5408\uff0c\u8f83\u5c11\u4e0e\u58f0\u97f3\u4e8b\u4ef6\u68c0\u6d4b\u3001DOA \u5b9a\u4f4d\u7edf\u4e00\u5efa\u6a21\u3002\u56e0\u6b64\uff0c\u672c\u6587\u5c06 SELD \u6269\u5c55\u4e3a 3D SELD\uff1a\u540c\u65f6\u8fdb\u884c\u58f0\u97f3\u4e8b\u4ef6\u68c0\u6d4b\u3001\u65b9\u5411\u5b9a\u4f4d\u548c\u8ddd\u79bb\u4f30\u8ba1\u3002<\/p>\n\n\n\n

\u6838\u5fc3\u8d21\u732e<\/h2>\n\n\n\n
    \n
  • \u63d0\u51fa 3D SELD \u4efb\u52a1\u5b9a\u4e49<\/strong>\uff1a\u5728\u539f\u6709 SELD \u7684\u4e8b\u4ef6\u7c7b\u522b\u4e0e DOA \u4f30\u8ba1\u57fa\u7840\u4e0a\uff0c\u52a0\u5165\u8ddd\u79bb\u4f30\u8ba1\uff0c\u4f7f\u7cfb\u7edf\u80fd\u591f\u4f30\u8ba1\u58f0\u6e90\u4e09\u7ef4\u4f4d\u7f6e\u3002<\/li>\n\n\n\n
  • \u63d0\u51fa Multi-ACCDDOA \u8868\u793a<\/strong>\uff1a\u5728 multi-ACCDOA \u7684\u57fa\u7840\u4e0a\uff0c\u628a\u539f\u672c\u7684\u4e09\u7ef4 DOA \u5411\u91cf\u6269\u5c55\u4e3a\u5305\u542b\u8ddd\u79bb\u7684\u8f93\u51fa\u8868\u793a\u3002<\/li>\n\n\n\n
  • \u6bd4\u8f83\u4e24\u79cd\u5efa\u6a21\u65b9\u5f0f<\/strong>\uff1a\u4e00\u79cd\u662f\u591a\u4efb\u52a1\u7ed3\u6784\uff0c\u5206\u522b\u8f93\u51fa SELD \u4e0e\u8ddd\u79bb\uff1b\u53e6\u4e00\u79cd\u662f\u5355\u4efb\u52a1\u7684 Multi-ACCDDOA \u7ed3\u6784\u3002<\/li>\n\n\n\n
  • \u6bd4\u8f83\u4e0d\u540c\u635f\u5931\u51fd\u6570<\/strong>\uff1a\u9488\u5bf9\u8ddd\u79bb\u4f30\u8ba1\u90e8\u5206\u5b9e\u9a8c MSE\u3001MAE\u3001MSPE\u3001MAPE\uff0c\u5206\u6790\u8ddd\u79bb\u635f\u5931\u5bf9 SELD \u6027\u80fd\u548c\u8ddd\u79bb\u8bef\u5dee\u7684\u5f71\u54cd\u3002<\/li>\n<\/ul>\n\n\n\n

    \u65b9\u6cd5<\/h2>\n\n\n\n
      \n
    • \u8f93\u5165\u7279\u5f81<\/strong>\uff1a\n
        \n
      • Ambisonics\/FOA\uff1a\u4f7f\u7528 4 \u901a\u9053\u5e45\u5ea6\u8c31\u8f6c\u6362\u4e3a 64 \u7ef4 Mel \u80fd\u91cf\uff0c\u5e76\u52a0\u5165 3 \u4e2a\u5f3a\u5ea6\u5411\u91cf\u7279\u5f81\uff0c\u603b\u8ba1 7 \u4e2a\u7279\u5f81\u901a\u9053\u3002<\/li>\n\n\n\n
      • Binaural\uff1a\u4f7f\u7528\u53cc\u8033\u5e73\u5747\u5e45\u5ea6\u8c31\uff0c\u5e76\u52a0\u5165 IPD \u7684 sine\/cosine \u8868\u793a\u548c ILD\uff0c\u603b\u8ba1 4 \u4e2a\u7279\u5f81\u901a\u9053\u3002<\/li>\n\n\n\n
      • STFT \u8bbe\u7f6e\uff1a40 ms Hamming \u7a97\uff0c50% overlap\uff1b\u6bcf\u4e2a\u8f93\u5165\u7247\u6bb5\u957f\u5ea6\u4e3a 250 \u5e27\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n
        \"\"<\/figure>\n\n\n\n
          \n
        • \u6a21\u578b\u7ed3\u6784<\/strong>\uff1a\n
            \n
          • \u4f7f\u7528 CRNN \u67b6\u6784\uff1a3 \u4e2a\u4e8c\u7ef4\u5377\u79ef\u5757 + 2 \u5c42\u53cc\u5411 GRU + 2 \u5c42 Multi-Head Attention\u3002<\/li>\n\n\n\n
          • \u6bcf\u4e2a\u5377\u79ef\u5757\u542b 128 \u4e2a\u5377\u79ef\u6838\u3001BatchNorm \u548c\u6c60\u5316\u3002<\/li>\n\n\n\n
          • \u6700\u540e\u901a\u8fc7\u5168\u8fde\u63a5\u5c42\u8f93\u51fa\u4e8b\u4ef6\u7c7b\u522b\u3001\u65b9\u5411\u548c\u8ddd\u79bb\u76f8\u5173\u4fe1\u606f\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n
              \n
            • \u65b9\u6cd5\u4e00\uff1aMulti-task<\/strong>\n
                \n
              • \u5206\u6210\u4e24\u4e2a\u8f93\u51fa\u5206\u652f\u3002<\/li>\n\n\n\n
              • \u7b2c\u4e00\u4e2a\u5206\u652f\u4f7f\u7528 ACCDOA \u8868\u793a\u5b8c\u6210 SELD\uff0c\u5373\u4e8b\u4ef6\u6d3b\u52a8\u4e0e DOA \u5411\u91cf\u8026\u5408\u3002<\/li>\n\n\n\n
              • \u7b2c\u4e8c\u4e2a\u5206\u652f\u5355\u72ec\u9884\u6d4b\u6bcf\u4e2a\u7c7b\u522b\u7684\u8ddd\u79bb\u3002<\/li>\n\n\n\n
              • \u4f18\u70b9\u662f\u7ed3\u6784\u76f4\u89c2\uff0c\u8ddd\u79bb\u4f30\u8ba1\u4e0e SELD \u53ef\u5206\u5f00\u4f18\u5316\u3002<\/li>\n\n\n\n
              • \u7f3a\u70b9\u662f\u4e0d\u80fd\u533a\u5206\u540c\u4e00\u7c7b\u522b\u4e2d\u591a\u4e2a\u91cd\u53e0\u58f0\u6e90\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n
              • \u65b9\u6cd5\u4e8c\uff1aMulti-ACCDDOA<\/strong>\n
                  \n
                • \u5c06 multi-ACCDOA \u6269\u5c55\u4e3a\u5305\u542b\u8ddd\u79bb\u7684\u8868\u793a\u3002<\/li>\n\n\n\n
                • \u8f93\u51fa\u5f62\u5f0f\u53ef\u4ee5\u7406\u89e3\u4e3a\uff1a\u6bcf\u4e2a track\u3001\u6bcf\u4e2a\u7c7b\u522b\u3001\u6bcf\u4e2a\u65f6\u95f4\u5e27\u540c\u65f6\u9884\u6d4b\u4e8b\u4ef6\u6d3b\u52a8\u3001DOA \u5411\u91cf\u548c\u8ddd\u79bb\u3002<\/li>\n\n\n\n
                • \u652f\u6301\u6700\u591a 3 \u4e2a\u91cd\u53e0\u58f0\u6e90\uff0c\u5e76\u53ef\u5904\u7406\u540c\u7c7b\u91cd\u53e0\u4e8b\u4ef6\u3002<\/li>\n\n\n\n
                • \u4f7f\u7528 ADPIT\uff0c\u5373 Auxiliary Duplicating Permutation Invariant Training\uff0c\u89e3\u51b3 track \u4e0e\u771f\u5b9e\u58f0\u6e90\u4e4b\u95f4\u7684\u6392\u5217\u5339\u914d\u95ee\u9898\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n

                  \u5b9e\u9a8c<\/h2>\n\n\n\n
                    \n
                  • \u6570\u636e\u96c6<\/strong>\uff1aSTARSS23\uff0cSony-TAU Realistic Spatial Soundscapes 2023\u3002<\/li>\n\n\n\n
                  • \u771f\u5b9e\u5f55\u97f3\u89c4\u6a21<\/strong>\uff1a7 \u5c0f\u65f6 22 \u5206\u949f\uff1b\u8bad\u7ec3\u96c6 90 \u4e2a clips\uff0c\u6d4b\u8bd5\u96c6 78 \u4e2a clips\u3002<\/li>\n\n\n\n
                  • \u7c7b\u522b\u6570<\/strong>\uff1a13 \u7c7b\uff0c\u5305\u62ec\u7537\u5973\u8bf4\u8bdd\u3001\u62cd\u624b\u3001\u7535\u8bdd\u3001\u7b11\u58f0\u3001\u5bb6\u5ead\u58f0\u97f3\u3001\u811a\u6b65\u3001\u95e8\u58f0\u3001\u97f3\u4e50\u3001\u4e50\u5668\u3001\u6c34\u9f99\u5934\u3001\u94c3\u58f0\u3001\u6572\u51fb\u7b49\u3002<\/li>\n\n\n\n
                  • \u91cd\u53e0\u60c5\u51b5<\/strong>\uff1a\u6700\u591a 3 \u4e2a\u540c\u65f6\u5b58\u5728\u7684\u58f0\u6e90\u3002<\/li>\n<\/ul>\n\n\n\n

                    \u7ed3\u679c<\/h2>\n\n\n\n

                    \u8bba\u6587\u5206\u522b\u5728\u00a0Ambisonics\/FOA \u97f3\u9891<\/strong>\u548c\u00a0Binaural \u53cc\u8033\u97f3\u9891<\/strong>\u4e0a\u8fdb\u884c\u4e86\u5b9e\u9a8c\uff0c\u5e76\u6bd4\u8f83\u4e86\u4e24\u7c7b\u65b9\u6cd5\uff1aMulti-task<\/strong>\u00a0\u548c\u00a0Multi-ACCDDOA<\/strong>\u3002\u8bc4\u4ef7\u6307\u6807\u5305\u62ec\u58f0\u97f3\u4e8b\u4ef6\u68c0\u6d4b\u9519\u8bef\u7387 ER\u3001F1 \u503c\u3001DOA \u5b9a\u4f4d\u8bef\u5dee\u3001\u5b9a\u4f4d\u53ec\u56de\u7387 Recall\uff0c\u4ee5\u53ca\u8ddd\u79bb\u4f30\u8ba1\u8bef\u5dee Dist. error\u3002<\/p>\n\n\n\n

                    \"\"<\/figure>\n\n\n\n

                    \u5728\u00a0Ambisonics<\/strong>\u00a0\u97f3\u9891\u4e0a\uff0c\u6574\u4f53\u6548\u679c\u6700\u597d\u7684\u662f\u00a0Multi-ACCDDOA + MSE<\/strong>\uff0c\u5176\u7ed3\u679c\u4e3a\uff1aER = 0.65\uff0cF1 = 44.2%\uff0cDOA error = 22.9\u00b0\uff0cRecall = 68.4%\uff0c\u8ddd\u79bb\u8bef\u5dee\u4e3a 0.92 m\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cMulti-ACCDDOA + MAE<\/strong>\u00a0\u867d\u7136\u53d6\u5f97\u4e86\u6700\u4f4e\u7684\u8ddd\u79bb\u8bef\u5dee 0.74 m\uff0c\u4f46 F1 \u4e0b\u964d\u5230 21.5%\uff0cRecall \u4e5f\u4e0b\u964d\u5230 19.1%\uff0c\u8bf4\u660e MAE \u4f1a\u660e\u663e\u635f\u5bb3\u4e8b\u4ef6\u68c0\u6d4b\u548c\u5b9a\u4f4d\u6027\u80fd\u3002<\/p>\n\n\n\n

                    \"\"<\/figure>\n\n\n\n

                    \u5728\u00a0Binaural<\/strong>\u00a0\u97f3\u9891\u4e0a\uff0c\u6574\u4f53\u6027\u80fd\u4f4e\u4e8e Ambisonics\u3002\u6700\u4f73\u7efc\u5408\u7ed3\u679c\u4ecd\u6765\u81ea\u00a0Multi-ACCDDOA + MSE<\/strong>\uff0c\u5176 ER = 0.87\uff0cF1 = 21.1%\uff0cDOA error = 39.7\u00b0\uff0cRecall = 48.0%\uff0c\u8ddd\u79bb\u8bef\u5dee\u4e3a 0.99 m\u3002Multi-ACCDDOA + MAE<\/strong>\u00a0\u7684\u8ddd\u79bb\u8bef\u5dee\u6700\u4f4e\uff0c\u4e3a 0.75 m\uff0c\u4f46 F1 \u4ec5\u4e3a 5.4%\uff0c\u68c0\u6d4b\u6027\u80fd\u4e25\u91cd\u4e0b\u964d\u3002<\/p>\n\n\n\n

                    \u7ed3\u8bba<\/h2>\n\n\n\n

                    \u672c\u6587\u8bc1\u660e\u4e86\u5728\u4e0d\u660e\u663e\u635f\u5bb3\u539f\u6709\u58f0\u97f3\u4e8b\u4ef6\u68c0\u6d4b\u548c DOA \u5b9a\u4f4d\u80fd\u529b\u7684\u60c5\u51b5\u4e0b\uff0c\u53ef\u4ee5\u628a\u8ddd\u79bb\u4f30\u8ba1\u52a0\u5165 SELD\uff0c\u5f62\u6210 3D SELD\u3002Multi-ACCDDOA \u6bd4\u7b80\u5355\u591a\u4efb\u52a1\u7ed3\u6784\u66f4\u6709\u4f18\u52bf\uff0c\u56e0\u4e3a\u5b83\u652f\u6301\u540c\u7c7b\u91cd\u53e0\u58f0\u6e90\uff0c\u5e76\u5728\u6574\u4f53 SELD \u6307\u6807\u4e0a\u8868\u73b0\u66f4\u597d\u3002\u4f46\u8ddd\u79bb\u4f30\u8ba1\u548c SELD \u7684\u6700\u4f18\u635f\u5931\u51fd\u6570\u5e76\u4e0d\u4e00\u81f4\uff1aMSE \u6709\u5229\u4e8e SELD\uff0cMAE \u6709\u5229\u4e8e\u8ddd\u79bb\u4f30\u8ba1\u3002\u672a\u6765\u5de5\u4f5c\u53ef\u4ee5\u8bbe\u8ba1\u6df7\u5408\u635f\u5931\u51fd\u6570\uff0c\u517c\u987e SELD \u7684\u7a33\u5b9a\u6027\u548c\u8ddd\u79bb\u4f30\u8ba1\u7cbe\u5ea6\uff1b\u4e5f\u53ef\u4ee5\u7ed3\u5408 Multi-ACCDDOA \u7684 track-wise \u8868\u793a\u4e0e\u591a\u4efb\u52a1\u8f93\u51fa\u7ed3\u6784\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

                    \u4f5c\u8005\uff1aDaniel Aleksander Krause, Archontis Politis, Annamar […]<\/p>\n","protected":false},"author":72,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-5207","post","type-post","status-publish","format-standard","hentry","category-events"],"_links":{"self":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/5207","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/users\/72"}],"replies":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/comments?post=5207"}],"version-history":[{"count":1,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/5207\/revisions"}],"predecessor-version":[{"id":5224,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/5207\/revisions\/5224"}],"wp:attachment":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/media?parent=5207"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/categories?post=5207"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/tags?post=5207"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}