{"id":4681,"date":"2024-09-18T16:37:03","date_gmt":"2024-09-18T08:37:03","guid":{"rendered":"https:\/\/nullthought.net\/?p=4681"},"modified":"2025-03-03T12:25:31","modified_gmt":"2025-03-03T04:25:31","slug":"masksr2%ef%bc%9a%e5%9f%ba%e4%ba%8emasksr%ef%bc%8c%e7%bb%93%e5%90%88%e8%af%ad%e4%b9%89%e7%9f%a5%e8%af%86%e8%92%b8%e9%a6%8f%e5%92%8c%e5%a3%b0%e5%ad%a6%e8%af%ad%e8%a8%80%e5%bb%ba%e6%a8%a1%e7%9a%84","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=4681","title":{"rendered":"MaskSR2\uff1a\u57fa\u4e8eMaskSR\uff0c\u7ed3\u5408\u8bed\u4e49\u77e5\u8bc6\u84b8\u998f\u548c\u58f0\u5b66\u8bed\u8a00\u5efa\u6a21\u7684\u5168\u9891\u5e26\u8bed\u97f3\u6062\u590d\u751f\u6210\u6846\u67b6"},"content":{"rendered":"\n<p>\u8bba\u6587<strong><a href=\"https:\/\/arxiv.org\/abs\/2409.09357\" target=\"_blank\" rel=\"noreferrer noopener\">Joint Semantic Knowledge Distillation and Masked Acoustic Modeling for Full-band Speech Restoration with Improved Intelligibility<\/a><\/strong>\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u8bed\u97f3\u6062\u590d\u751f\u6210\u6846\u67b6MaskSR2\uff0c\u4e3b\u8981\u76ee\u6807\u662f\u63d0\u9ad8\u6062\u590d\u8bed\u97f3\u7684\u6e05\u6670\u5ea6\u548c\u8d28\u91cf\uff0c\u5c24\u5176\u662f\u5728\u5168\u9891\u5e26\uff0844.1kHz\uff09\u6761\u4ef6\u4e0b\u3002\u8be5\u6846\u67b6\u57fa\u4e8e\u5148\u524d\u7684MaskSR\u6a21\u578b\uff0c\u7ed3\u5408\u4e86<strong>\u8bed\u4e49\u77e5\u8bc6\u84b8\u998f<\/strong>\uff08<strong>Semantic Knowledge Distillation<\/strong>\uff09\u548c<strong>\u63a9\u7801\u58f0\u5b66\u5efa\u6a21<\/strong>\uff08<strong>Masked Acoustic Modeling<\/strong>\uff09\u6765\u9884\u6d4b\u76ee\u6807\u8bed\u97f3\u7684\u58f0\u5b66\u7ec6\u8282\uff0c\u4ece\u800c\u6539\u5584\u8bed\u97f3\u7684\u53ef\u61c2\u5ea6\uff08\u7528\u8bcd\u9519\u8bef\u7387WER\u8861\u91cf\uff09\u548c\u6574\u4f53\u8d28\u91cf\u3002<\/p>\n\n\n\n<p>\u8bba\u6587\u4f5c\u8005\u4e3aXiaoyu Liu, Xu Li, Joan Serr\u00e0, Santiago Pascual\uff0c\u5747\u6765\u81eaDolby Laboratories\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"912\" height=\"347\" src=\"https:\/\/nullthought.net\/wp-content\/uploads\/2024\/09\/image-8.png\" alt=\"\" class=\"wp-image-4682\" srcset=\"https:\/\/nullthought.net\/wp-content\/uploads\/2024\/09\/image-8.png 912w, https:\/\/nullthought.net\/wp-content\/uploads\/2024\/09\/image-8-300x114.png 300w, https:\/\/nullthought.net\/wp-content\/uploads\/2024\/09\/image-8-768x292.png 768w\" sizes=\"auto, (max-width: 912px) 100vw, 912px\" \/><figcaption class=\"wp-element-caption\"><strong><a href=\"https:\/\/arxiv.org\/abs\/2409.09357\" target=\"_blank\" rel=\"noreferrer noopener\">Joint Semantic Knowledge Distillation and Masked Acoustic Modeling for Full-band Speech Restoration with Improved Intelligibility<\/a><\/strong><\/figcaption><\/figure>\n\n\n\n<p>\u8bba\u6587\u5185\u5bb9\u6982\u8981\u5982\u4e0b\uff1a<\/p>\n\n\n\n<p><strong>\u8bed\u97f3\u6062\u590d\uff08Speech Restoration, SR\uff09<\/strong> \u662f\u4ece\u53d7\u635f\u7684\u8bed\u97f3\u4fe1\u53f7\u4e2d\u6062\u590d\u51fa\u9ad8\u8d28\u91cf\u7684\u3001\u5168\u9891\u5e26\uff0844.1kHz\uff09\u8bed\u97f3\u3002\u76f8\u6bd4\u4f20\u7edf\u7684\u8bed\u97f3\u589e\u5f3a\uff08Speech Enhancement, SE\uff09\u4e3b\u8981\u89e3\u51b3\u53bb\u566a\u548c\u53bb\u6df7\u54cd\u95ee\u9898\uff0c\u8bed\u97f3\u6062\u590d\u66f4\u5177\u6311\u6218\u6027\uff0c\u56e0\u4e3a\u5b83\u8fd8\u8981\u5904\u7406\u5e26\u5bbd\u6269\u5c55\u3001\u4e22\u5305\u7b49\u95ee\u9898\u3002\u5c3d\u7ba1\u73b0\u6709\u7684\u751f\u6210\u6a21\u578b\u5982MaskSR\u5728\u6062\u590d\u8bed\u97f3\u7684\u611f\u77e5\u8d28\u91cf\u4e0a\u53d6\u5f97\u4e86\u4e0d\u9519\u7684\u7ed3\u679c\uff0c\u4f46\u5374\u65e0\u6cd5\u663e\u8457\u63d0\u9ad8\u8bed\u97f3\u7684\u53ef\u61c2\u5ea6\uff08Word Error Rate, WER\uff09\uff0c\u751a\u81f3\u5728\u67d0\u4e9b\u60c5\u51b5\u4e0b\uff0c\u6062\u590d\u540e\u7684\u8bed\u97f3WER\u6bd4\u53d7\u635f\u8bed\u97f3\u66f4\u9ad8\u3002\u8fd9\u662f\u56e0\u4e3a\u73b0\u6709\u6a21\u578b\u5728\u53bb\u9664\u5931\u771f\u7684\u8fc7\u7a0b\u4e2d\u53ef\u80fd\u6539\u53d8\u8bed\u97f3\u4e2d\u7684\u97f3\u97f5\u4fe1\u606f\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u89e3\u51b3\u4ee5\u4e0a\u95ee\u9898\uff0c\u8bba\u6587\u63d0\u51fa\u4e86<strong>MaskSR2<\/strong>\uff0c\u5728MaskSR\u7684\u57fa\u7840\u4e0a\uff0c\u901a\u8fc7\u5f15\u5165\u8bed\u4e49\u77e5\u8bc6\u84b8\u998f\uff08Semantic Knowledge Distillation, KD\uff09\u673a\u5236\uff0c\u663e\u8457\u6539\u5584\u4e86\u8bed\u97f3\u7684\u53ef\u61c2\u5ea6\uff0c\u540c\u65f6\u4fdd\u6301\u4e86\u9ad8\u8d28\u91cf\u3002MaskSR2\u7684\u6838\u5fc3\u521b\u65b0\u5728\u4e8e\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u65e0\u9700\u8f6c\u5f55\u6570\u636e<\/strong>\uff1a\u901a\u8fc7\u4f7f\u7528\u81ea\u76d1\u7763\u5b66\u4e60\u7684\u9884\u8bad\u7ec3\u6a21\u578bHuBERT\u6765\u63d0\u53d6\u8bed\u97f3\u7684\u8bed\u4e49\u8868\u793a\uff0c\u907f\u514d\u4e86\u4f9d\u8d56\u5927\u578b\u8f6c\u5f55\u8bed\u6599\u3002<\/li>\n\n\n\n<li><strong>\u5728\u63a8\u7406\u9636\u6bb5\u8ba1\u7b97\u6210\u672c\u4e0d\u53d8<\/strong>\uff1a\u8bad\u7ec3\u65f6\u4f7f\u7528HuBERT\u4f5c\u4e3a\u6559\u5e08\u6a21\u578b\uff0c\u63a8\u7406\u65f6\u820d\u5f03\u8be5\u6a21\u578b\uff0c\u56e0\u800c\u5728\u63a8\u7406\u65f6\u7684\u8ba1\u7b97\u5f00\u9500\u4e0eMaskSR\u76f8\u540c\u3002<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">1. \u6280\u672f\u65b9\u6cd5\u8be6\u89e3<\/h4>\n\n\n\n<h5 class=\"wp-block-heading\">1.1 MaskSR2\u67b6\u6784<\/h5>\n\n\n\n<p>\u8bba\u6587\u7684\u521b\u65b0\u6a21\u578bMaskSR2\u901a\u8fc7\u4e24\u4e2a\u4e3b\u8981\u6a21\u5757\u6765\u5b9e\u73b0\u8bed\u97f3\u6062\u590d\uff1a<strong>\u8bed\u97f3\u7f16\u7801\u5668\uff08Speech Encoder\uff09\u548c\u751f\u6210\u6a21\u578b\uff08Generative Model\uff09<\/strong>\u3002<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u8bed\u97f3\u7f16\u7801\u5668<\/strong>\uff1a\u8bed\u97f3\u7f16\u7801\u5668\u7684\u4efb\u52a1\u662f\u4ece\u53d7\u635f\u7684\u8bed\u97f3\u4fe1\u53f7\u4e2d\u63d0\u53d6\u5176\u8bed\u4e49\u4fe1\u606f\u3002\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u8f93\u5165\u7684\u8bed\u97f3\u4fe1\u53f7\u5148\u7ecf\u8fc7\u77ed\u65f6\u5085\u91cc\u53f6\u53d8\u6362\uff08STFT\uff09\uff0c\u7136\u540e\u901a\u8fc7\u6279\u91cf\u5f52\u4e00\u5316\u5c42\u548c\u5168\u8fde\u63a5\u5c42\u8f6c\u5316\u4e3a\u56fa\u5b9a\u7ef4\u5ea6\u7684\u7279\u5f81\u8868\u793a\u3002\u968f\u540e\uff0c\u8fd9\u4e9b\u7279\u5f81\u88ab\u8f93\u5165\u5230\u4e00\u7cfb\u5217\u81ea\u6ce8\u610f\u529b\u7684Transformer\u5757\u4e2d\uff0c\u751f\u6210\u8bed\u97f3\u7684\u6f5c\u5728\u8bed\u4e49\u5d4c\u5165\u3002\u901a\u8fc7\u5f15\u5165<strong>\u8bed\u4e49\u77e5\u8bc6\u84b8\u998f<\/strong>\u673a\u5236\uff0cMaskSR2\u4f7f\u7528\u4e00\u4e2a\u9884\u8bad\u7ec3\u7684HuBERT\u6a21\u578b\u4f5c\u4e3a\u201c\u6559\u5e08\u6a21\u578b\u201d\uff0c\u4e3a\u8bed\u97f3\u7f16\u7801\u5668\u63d0\u4f9b\u76ee\u6807\u8bed\u97f3\u7684\u8bed\u4e49\u8868\u793a\uff0c\u4ece\u800c\u65e0\u9700\u4f9d\u8d56\u8f6c\u5f55\u6587\u672c\u6765\u6307\u5bfc\u6a21\u578b\u8bad\u7ec3\u3002<\/li>\n\n\n\n<li><strong>\u751f\u6210\u6a21\u578b<\/strong>\uff1a\u751f\u6210\u6a21\u578b\u57fa\u4e8e\u63a9\u7801\u751f\u6210\u6280\u672fMaskGIT\u3002\u76ee\u6807\u8bed\u97f3\u7684\u58f0\u5b66\u6807\u8bb0\uff08acoustic tokens\uff09\u901a\u8fc7\u9884\u8bad\u7ec3\u7684DAC\u6a21\u578b\u8fdb\u884c\u91cf\u5316\uff0c\u751f\u6210\u4e00\u7ec4\u79bb\u6563\u7684\u6807\u8bb0\u3002\u8fd9\u4e9b\u6807\u8bb0\u7f16\u7801\u4e86\u76ee\u6807\u8bed\u97f3\u7684\u7ec6\u8282\u58f0\u5b66\u4fe1\u606f\u3002\u6a21\u578b\u901a\u8fc7\u63a9\u7801\u673a\u5236\uff0c\u9010\u6b65\u586b\u8865\u88ab\u63a9\u76d6\u7684\u58f0\u5b66\u6807\u8bb0\uff0c\u6700\u7ec8\u751f\u6210\u76ee\u6807\u8bed\u97f3\u6ce2\u5f62\u3002<\/li>\n<\/ol>\n\n\n\n<h5 class=\"wp-block-heading\">1.2 \u8bed\u4e49\u77e5\u8bc6\u84b8\u998f<\/h5>\n\n\n\n<p>MaskSR2\u5728\u8bed\u97f3\u7f16\u7801\u5668\u4e2d\u5f15\u5165\u4e86\u8bed\u4e49\u77e5\u8bc6\u84b8\u998f\u673a\u5236\uff0c\u5373\u4f7f\u7528\u6765\u81ea\u9884\u8bad\u7ec3HuBERT\u6a21\u578b\u7684\u8bed\u4e49\u8868\u793a\u6307\u5bfc\u7f16\u7801\u5668\u5b66\u4e60\u3002\u8fd9\u4e2a\u84b8\u998f\u8fc7\u7a0b\u901a\u8fc7\u4e09\u4e2a\u4e0d\u540c\u7684\u7279\u5f81\u9009\u53d6\u7b56\u7565\u5b9e\u73b0\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>L9-K500<\/strong>\uff1a\u5c06HuBERT\u7b2c9\u5c42\u7684\u7279\u5f81\u8fdb\u884ck-means\u805a\u7c7b\uff0c\u751f\u6210500\u4e2a\u79bb\u6563\u7684\u4f2a\u97f3\u7d20\u7c7b\u3002<\/li>\n\n\n\n<li><strong>L9-feature<\/strong>\uff1a\u76f4\u63a5\u4f7f\u7528\u7b2c9\u5c42\u7684\u8fde\u7eed\u7279\u5f81\uff0c\u8fd9\u4e9b\u7279\u5f81\u5305\u542b\u4e86\u4e30\u5bcc\u7684\u97f3\u97f5\u4fe1\u606f\u548c\u5176\u4ed6\u4e0e\u8bed\u97f3\u6062\u590d\u4efb\u52a1\u76f8\u5173\u7684\u7279\u5f81\uff0c\u5982\u8bf4\u8bdd\u4eba\u5dee\u5f02\u548c\u97f5\u5f8b\u4fe1\u606f\u3002<\/li>\n\n\n\n<li><strong>Avg-feature<\/strong>\uff1a\u4f7f\u7528HuBERT\u6a21\u578b\u6240\u670912\u5c42\u7684\u5e73\u5747\u7279\u5f81\uff0c\u8fd9\u6837\u53ef\u4ee5\u5229\u7528\u6bcf\u5c42\u7279\u5f81\u5728\u4e0d\u540c\u4efb\u52a1\u4e0a\u7684\u4f18\u52bf\uff0c\u53ef\u80fd\u5bf9\u751f\u6210\u8bed\u97f3\u7684\u6574\u4f53\u8d28\u91cf\u66f4\u6709\u5e2e\u52a9\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4e3a\u4e86\u786e\u4fdd\u8bad\u7ec3\u7684\u7a33\u5b9a\u6027\uff0c\u76ee\u6807\u7279\u5f81\u5728\u6bcf\u4e2a\u8bad\u7ec3\u6837\u672c\u4e0a\u90fd\u8fdb\u884c\u4e86\u5747\u503c\u548c\u65b9\u5dee\u5f52\u4e00\u5316\u3002\u901a\u8fc7\u4e0e\u4f20\u7edf\u7684\u9891\u8c31\u76ee\u6807\u5bf9\u6bd4\u5b9e\u9a8c\uff0c\u8bba\u6587\u8868\u660e\uff0c\u4f7f\u7528HuBERT\u7684\u8bed\u4e49\u8868\u793a\u8fdb\u884c\u84b8\u998f\u76f8\u6bd4\u4e8e\u76f4\u63a5\u56de\u5f52\u5230\u9891\u8c31\u7279\u5f81\uff08\u5982STFT\uff09\u663e\u8457\u6539\u5584\u4e86\u8bed\u97f3\u7684\u53ef\u61c2\u5ea6\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">1.3 \u63a9\u7801\u58f0\u5b66\u5efa\u6a21<\/h5>\n\n\n\n<p>\u751f\u6210\u6a21\u578b\u4f7f\u7528MaskGIT\u673a\u5236\u8fdb\u884c\u63a9\u7801\u58f0\u5b66\u5efa\u6a21\u3002\u76ee\u6807\u8bed\u97f3\u7684\u58f0\u5b66\u6807\u8bb0\u901a\u8fc7\u9884\u8bad\u7ec3\u7684DAC\u6a21\u578b\u8fdb\u884c\u91cf\u5316\u3002\u8bad\u7ec3\u65f6\uff0c\u90e8\u5206\u58f0\u5b66\u6807\u8bb0\u88ab\u968f\u673a\u63a9\u76d6\uff0c\u751f\u6210\u6a21\u578b\u901a\u8fc7\u8fd9\u4e9b\u63a9\u7801\u4f4d\u7f6e\u6765\u9884\u6d4b\u76ee\u6807\u8bed\u97f3\u7684\u58f0\u5b66\u6807\u8bb0\u3002\u63a8\u7406\u9636\u6bb5\uff0c\u6a21\u578b\u901a\u8fc7\u8fed\u4ee3\u751f\u6210\u76ee\u6807\u58f0\u5b66\u6807\u8bb0\uff0c\u6700\u7ec8\u751f\u6210\u8bed\u97f3\u6ce2\u5f62\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">1.4 \u63a8\u7406\u4f18\u5316<\/h5>\n\n\n\n<p>\u5728\u63a8\u7406\u9636\u6bb5\uff0cMaskSR2\u901a\u8fc7\u91c7\u6837\u673a\u5236\u751f\u6210\u58f0\u5b66\u6807\u8bb0\uff0c\u5e76\u5728\u6bcf\u6b21\u8fed\u4ee3\u4e2d\u91cd\u65b0\u63a9\u76d6\u4f4e\u7f6e\u4fe1\u5ea6\u7684\u6807\u8bb0\u3002\u8fd9\u79cd\u8fed\u4ee3\u751f\u6210\u7684\u8fc7\u7a0b\u80fd\u591f\u6709\u6548\u6539\u5584\u76ee\u6807\u8bed\u97f3\u7684\u8d28\u91cf\u3002\u6b64\u5916\uff0cMaskSR2\u901a\u8fc7<strong>Classifier-free guidance<\/strong>\u7b56\u7565\u6765\u8fdb\u4e00\u6b65\u964d\u4f4eWER\uff0c\u8fd9\u4e2a\u7b56\u7565\u901a\u8fc7\u5728\u8bad\u7ec3\u65f6\u968f\u673a\u66ff\u6362\u7f16\u7801\u5668\u8f93\u51fa\uff0c\u5728\u63a8\u7406\u65f6\u4ee5\u66f4\u5927\u6743\u91cd\u6761\u4ef6\u5316\u751f\u6210\u8fc7\u7a0b\uff0c\u786e\u4fdd\u751f\u6210\u7684\u8bed\u97f3\u80fd\u591f\u66f4\u597d\u5730\u4fdd\u7559\u539f\u59cb\u97f3\u7d20\u4fe1\u606f\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">2. \u5b9e\u9a8c\u8bbe\u8ba1\u4e0e\u7ed3\u679c<\/h4>\n\n\n\n<h5 class=\"wp-block-heading\">2.1 \u5b9e\u9a8c\u6570\u636e<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u8bad\u7ec3\u96c6<\/strong>\uff1a\u4f7f\u7528\u4e86\u5305\u62ecDNS Challenge\u3001VCTK\u548cAISHELL-1\u7b49\u591a\u79cd\u6e05\u6670\u8bed\u97f3\u6570\u636e\uff0c\u603b\u8ba1800\u5c0f\u65f6\uff0c\u5e76\u4f7f\u7528181\u5c0f\u65f6\u7684\u566a\u58f0\u6570\u636e\u548c60,000\u4e2a\u623f\u95f4\u8109\u51b2\u54cd\u5e94\u6765\u6a21\u62df\u4e0d\u540c\u7684\u5931\u771f\u60c5\u51b5\u3002<\/li>\n\n\n\n<li><strong>\u6d4b\u8bd5\u96c6<\/strong>\uff1a\u4e3a\u4e86\u8bc4\u4f30\u6a21\u578b\u6027\u80fd\uff0c\u8bba\u6587\u4f7f\u7528\u4e86\u81ea\u751f\u6210\u768444.1kHz\u5168\u9891\u5e26\u6d4b\u8bd5\u96c6\u548c16kHz\u5bbd\u9891\u5e26\u6d4b\u8bd5\u96c6\uff0c\u6d4b\u8bd5\u6837\u672c\u4e2d\u5305\u542b\u566a\u58f0\u3001\u6df7\u54cd\u3001\u5e26\u5bbd\u9650\u5236\u548c\u526a\u5207\u7b49\u5931\u771f\u60c5\u51b5\u3002<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">2.2 \u8bc4\u4ef7\u6307\u6807<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>DNSMOS<\/strong> \u548c <strong>SESQA<\/strong> \u4f5c\u4e3a\u53c2\u8003\u81ea\u7531\u611f\u77e5\u8d28\u91cf\u7684\u8bc4\u4f30\u6307\u6807\uff0c\u7528\u4e8e\u8bc4\u4f30\u751f\u6210\u6a21\u578b\u7684\u6574\u4f53\u611f\u77e5\u8d28\u91cf\u3002<\/li>\n\n\n\n<li><strong>WER<\/strong>\uff08\u8bcd\u9519\u8bef\u7387\uff09\u7528\u4e8e\u8bc4\u4f30\u8bed\u97f3\u7684\u53ef\u61c2\u5ea6\uff0c\u901a\u8fc7ASR\u7cfb\u7edf\u6765\u6d4b\u91cf\u3002<\/li>\n\n\n\n<li><strong>Speaker Similarity<\/strong>\uff1a\u901a\u8fc7\u4f59\u5f26\u76f8\u4f3c\u5ea6\u6765\u8861\u91cf\u751f\u6210\u8bed\u97f3\u4e0e\u76ee\u6807\u8bed\u97f3\u7684\u8bf4\u8bdd\u4eba\u76f8\u4f3c\u5ea6\u3002<\/li>\n\n\n\n<li><strong>\u4e3b\u89c2\u542c\u89c9\u6d4b\u8bd5<\/strong>\uff1a\u901a\u8fc713\u540d\u4e13\u5bb6\u542c\u4f17\u5bf9\u751f\u6210\u8bed\u97f3\u7684\u4e3b\u89c2\u8bc4\u5206\u8fdb\u884c\u9a8c\u8bc1\uff0c\u7efc\u5408\u8bc4\u4f30\u8bed\u97f3\u7684\u611f\u77e5\u8d28\u91cf\u3002<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">2.3 \u5b9e\u9a8c\u7ed3\u679c<\/h5>\n\n\n\n<p>\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cMaskSR2\u663e\u8457\u63d0\u9ad8\u4e86\u751f\u6210\u8bed\u97f3\u7684\u53ef\u61c2\u5ea6\u548c\u611f\u77e5\u8d28\u91cf\u3002\u5177\u4f53\u5730\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5728VCTK\u6d4b\u8bd5\u96c6\u4e0a\uff0cMaskSR2\u76f8\u8f83\u4e8eMaskSR\u5728\u4fdd\u6301\u76f8\u540c\u63a8\u7406\u65f6\u95f4\u548c\u6a21\u578b\u5bb9\u91cf\u7684\u60c5\u51b5\u4e0b\uff0cWER\u964d\u4f4e\u4e8637.9%\u3002<\/li>\n\n\n\n<li>\u4e3b\u89c2\u542c\u89c9\u6d4b\u8bd5\u663e\u793a\uff0cMaskSR2\u751f\u6210\u7684\u8bed\u97f3\u8d28\u91cf\u63a5\u8fd1\u76ee\u6807\u8bed\u97f3\uff0c\u5e76\u663e\u8457\u4f18\u4e8e\u5176\u4ed6\u751f\u6210\u6a21\u578b\u3002<\/li>\n\n\n\n<li>\u5728\u66f4\u5177\u6311\u6218\u6027\u7684LibriSpeech\u6d4b\u8bd5\u96c6\u4e0a\uff0cMaskSR2\u7684WER\u964d\u4f4e\u4e8618.9%\uff0c\u8fdb\u4e00\u6b65\u8bc1\u660e\u4e86\u5176\u8bed\u4e49\u77e5\u8bc6\u84b8\u998f\u5bf9\u8bed\u97f3\u751f\u6210\u53ef\u61c2\u5ea6\u7684\u91cd\u8981\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">3. \u672a\u6765\u5de5\u4f5c\u65b9\u5411<\/h4>\n\n\n\n<p>\u8bba\u6587\u63d0\u5230\uff0c\u672a\u6765\u53ef\u80fd\u63a2\u7d22\u66f4\u5f3a\u5927\u7684\u81ea\u76d1\u7763\u5b66\u4e60\u6a21\u578b\u548c\u591a\u4efb\u52a1\u8bed\u97f3\u7f16\u7801\u5668\u6765\u8fdb\u4e00\u6b65\u63d0\u5347\u6027\u80fd\u3002\u4f8b\u5982\uff0c\u8003\u8651\u4f7f\u7528WavLM\u7b49\u66f4\u5f3a\u5927\u7684\u6a21\u578b\uff0c\u6216\u901a\u8fc7\u591a\u4e2a\u6559\u5e08\u6a21\u578b\u7684\u8054\u5408\u5b66\u4e60\u6765\u63d0\u5347\u751f\u6210\u8bed\u97f3\u7684\u8d28\u91cf\u548c\u53ef\u61c2\u5ea6\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u8bba\u6587Joint Semantic Knowledge Distillation and Masked Acou [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[8],"tags":[39,95,97],"class_list":["post-4681","post","type-post","status-publish","format-standard","hentry","category-tech","tag-ai","tag-transformer","tag-distillation"],"rttpg_featured_image_url":null,"rttpg_author":{"display_name":"NullThought","author_link":"https:\/\/nullthought.net\/?author=1"},"rttpg_comment":0,"rttpg_category":"<a href=\"https:\/\/nullthought.net\/?cat=8\" rel=\"category\">Tech<\/a>","rttpg_excerpt":"\u8bba\u6587Joint Semantic Knowledge Distillation and Masked Acou&hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4681","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4681"}],"version-history":[{"count":1,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4681\/revisions"}],"predecessor-version":[{"id":4683,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4681\/revisions\/4683"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4681"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4681"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4681"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}