{"id":38134,"date":"2025-02-10T21:01:39","date_gmt":"2025-02-10T13:01:39","guid":{"rendered":"https:\/\/17aitech.com\/?p=38134"},"modified":"2025-02-10T21:01:39","modified_gmt":"2025-02-10T13:01:39","slug":"iclr-2025%ef%bd%9c%e5%b0%8f%e7%b1%b3%e6%96%b0%e4%b8%80%e4%bb%a3kaldi%e8%af%ad%e9%9f%b3%e8%af%86%e5%88%ab%e7%ae%97%e6%b3%95cr-ctc%ef%bc%8c%e7%ba%afctc%e6%80%a7%e8%83%bd%e5%ae%9e%e7%8e%b0sota","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=38134","title":{"rendered":"ICLR 2025\uff5c\u5c0f\u7c73\u65b0\u4e00\u4ee3Kaldi\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5CR-CTC\uff0c\u7eafCTC\u6027\u80fd\u5b9e\u73b0SOTA"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-07-5\" target=\"_blank\">ICLR 2025\uff5c\u5c0f\u7c73\u65b0\u4e00\u4ee3Kaldi\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5CR-CTC\uff0c\u7eafCTC\u6027\u80fd\u5b9e\u73b0SOTA<\/a><\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1daf4275c3cb8450850a3f96506b97a2.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1daf4275c3cb8450850a3f96506b97a2.png\"><\/a><\/p>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"166\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<p>AIxiv\u4e13\u680f\u662f\u673a\u5668\u4e4b\u5fc3\u53d1\u5e03\u5b66\u672f\u3001\u6280\u672f\u5185\u5bb9\u7684\u680f\u76ee\u3002\u8fc7\u53bb\u6570\u5e74\uff0c\u673a\u5668\u4e4b\u5fc3AIxiv\u4e13\u680f\u63a5\u6536\u62a5\u9053\u4e862000\u591a\u7bc7\u5185\u5bb9\uff0c\u8986\u76d6\u5168\u7403\u5404\u5927\u9ad8\u6821\u4e0e\u4f01\u4e1a\u7684\u9876\u7ea7\u5b9e\u9a8c\u5ba4\uff0c\u6709\u6548\u4fc3\u8fdb\u4e86\u5b66\u672f\u4ea4\u6d41\u4e0e\u4f20\u64ad\u3002\u5982\u679c\u60a8\u6709\u4f18\u79c0\u7684\u5de5\u4f5c\u60f3\u8981\u5206\u4eab\uff0c\u6b22\u8fce\u6295\u7a3f\u6216\u8005\u8054\u7cfb\u62a5\u9053\u3002\u6295\u7a3f\u90ae\u7bb1\uff1aliyazhou@jiqizhixin.com\uff1bzhaoyunfeng@jiqizhixin.com<\/p>\n<\/section>\n<\/blockquote>\n<p><strong>\u65b0\u4e00\u4ee3 Kaldi \u56e2\u961f\u662f\u7531 Kaldi \u4e4b\u7236\u3001IEEE fellow\u3001\u5c0f\u7c73\u96c6\u56e2\u9996\u5e2d\u8bed\u97f3\u79d1\u5b66\u5bb6 Daniel Povey \u9886\u8854\u7684\u56e2\u961f\uff0c\u4e13\u6ce8\u4e8e\u5f00\u6e90\u8bed\u97f3\u57fa\u7840\u5f15\u64ce\u7814\u53d1\uff0c\u4ece\u795e\u7ecf\u7f51\u7edc\u58f0\u5b66\u7f16\u7801\u5668\u3001\u635f\u5931\u51fd\u6570\u3001\u4f18\u5316\u5668\u548c\u89e3\u7801\u5668\u7b49\u5404\u65b9\u9762\u91cd\u6784\u8bed\u97f3\u6280\u672f\u94fe\u8def\uff0c\u65e8\u5728\u63d0\u9ad8\u667a\u80fd\u8bed\u97f3\u4efb\u52a1\u7684\u51c6\u786e\u7387\u548c\u6548\u7387\u3002<\/strong><\/p>\n<p><strong>\u76ee\u524d\uff0c\u65b0\u4e00\u4ee3 Kaldi \u9879\u76ee \uff08https:\/\/github.com\/k2-fsa\uff09\u4e3b\u8981\u7531\u56db\u4e2a\u5b50\u9879\u76ee\u6784\u6210\uff1a\u6838\u5fc3\u7b97\u6cd5\u5e93 k2\u3001\u901a\u7528\u8bed\u97f3\u6570\u636e\u5904\u7406\u5de5\u5177\u5305 Lhotse\u3001\u89e3\u51b3\u65b9\u6848\u96c6\u5408 Icefall \u4ee5\u53ca\u670d\u52a1\u7aef\u5f15\u64ce Sherpa\uff0c\u65b9\u4fbf\u5f00\u53d1\u8005\u8f7b\u677e\u8bad\u7ec3\u3001\u90e8\u7f72\u81ea\u5df1\u7684\u667a\u80fd\u8bed\u97f3\u6a21\u578b\u3002<\/strong><\/p>\n<p>\u8fd1\u65e5\uff0c\u5c0f\u7c73\u96c6\u56e2\u65b0\u4e00\u4ee3 Kaldi \u56e2\u961f\u5173\u4e8e\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5\u7684\u8bba\u6587\u300aCR-CTC: Consistency regularization on CTC for improved speech recognition\u300b\u88ab ICLR 2025 \u63a5\u6536\u3002<\/p>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-bc882892d3b37c4e055cb2a053148022.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-bc882892d3b37c4e055cb2a053148022.png\"><\/a><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/pdf\/2410.05101<\/section>\n<\/li>\n<li>\n<section>\u8bba\u6587\u4ee3\u7801\uff1ahttps:\/\/github.com\/k2-fsa\/icefall\/pull\/1766\uff08\u5df2 merge \u8fdb icefall \u6846\u67b6\uff09<\/section>\n<\/li>\n<\/ul>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-a50605e6b1885b0c22f7cd6c0d3c6540.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-a50605e6b1885b0c22f7cd6c0d3c6540.png\"><\/a><\/section>\n<section><strong>\u6458\u8981<\/strong><\/section>\n<section><\/section>\n<section>\u4e3b\u6d41\u7684\u81ea\u52a8\u8bed\u97f3\u8bc6\u522b\uff08ASR\uff09\u6a21\u578b\u5305\u62ec CTC [1]\u3001transducer [2] \u548c\u6df7\u5408\u7cfb\u7edf CTC\/AED [3]\u3002CTC \u662f\u5176\u4e2d\u6700\u7b80\u5355\u3001\u6700\u4fbf\u4e8e\u90e8\u7f72\u7684\u65b9\u6cd5\uff0c\u4f46\u7531\u4e8e\u5b83\u7684\u6027\u80fd\u901a\u5e38\u660e\u663e\u843d\u540e\u4e8e Transducer \u548c CTC\/AED\uff0c\u8fd9\u9650\u5236\u4e86\u5b83\u7684\u5b9e\u9645\u5e94\u7528\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u6b64\uff0c\u65b0\u4e00\u4ee3 Kaldi \u56e2\u961f\u63d0\u51fa\u4e86 Consistency-Regularized CTC (CR-CTC)\uff0c\u53ef\u4ee5\u8ba9\u7eaf CTC \u6a21\u578b\u7684\u8bc6\u522b\u6027\u80fd\u6bd4\u80a9 Transducer \u548c CTC\/AED\u3002CR-CTC \u5728\u591a\u4e2a\u4e3b\u6d41\u7684 ASR \u6570\u636e\u96c6\uff0c\u5305\u62ec LibriSpeech\u3001Aishell-1\u3001GigaSpeech \u7b49\u6570\u636e\u96c6\u4e0a\uff0c\u53d6\u5f97\u65b0\u7684 SOTA \u7ed3\u679c\uff08\u4e0d\u4f9d\u8d56\u5916\u90e8\u8bad\u7ec3\u6570\u636e\u548c\u5916\u90e8\u8bed\u8a00\u6a21\u578b\uff09\u3002<\/section>\n<section><\/section>\n<section>\u4f8b\u5982\uff0c\u5728 LibriSpeech \u6570\u636e\u96c6\u4e0a\u8bad\u7ec3 Zipformer-L\uff0c\u6807\u51c6 CTC \u7684 WER \u4e3a 2.5\/5.72\uff0cCTC\/AED \u7684 WER \u4e3a 2.09\/4.59, Pruned Transducer \u7684 WER \u4e3a 2.00\/4.38\uff1bCR-CTC \u7684 WER \u4e3a 2.02\/4.35\uff1bCTC\/AED \u548c Pruned Transducer \u6302\u4e0a CR-CTC \u8054\u5408\u8bad\u7ec3\u540e\uff0cWER \u53ef\u5206\u522b\u8fdb\u4e00\u6b65\u964d\u4f4e\u5230 1.96\/4.08 \u548c 1.88\/3.95\u3002<\/section>\n<section><\/section>\n<section><strong>\u65b9\u6cd5\u5b9e\u73b0<\/strong><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f73caeb65e737fdaa1aae889e86b0ed5.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f73caeb65e737fdaa1aae889e86b0ed5.png\"><\/a><\/section>\n<section>\u5982 Figure 1 \u6240\u793a\uff0cCR-CTC \u65b9\u6cd5\u975e\u5e38\u7b80\u5355\uff0c\u5148\u4ece\u540c\u4e00\u4e2a\u8f93\u5165 Mel-spectrogram \u00a0x \u5f97\u5230\u4e24\u4e2a\u4e0d\u540c\u7684 augmented views <a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-86e2e49374e77b0afe7a39b9d72e48ad.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-86e2e49374e77b0afe7a39b9d72e48ad.png\"><\/a> \u548c<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-38fcd8aa28267fb4fc669200719111c5.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-38fcd8aa28267fb4fc669200719111c5.png\"><\/a>\uff0c\u5206\u522b\u8f93\u5165\u53c2\u6570\u5171\u4eab\u7684 encoder \u6a21\u578b f\uff0c\u5f97\u5230\u5bf9\u5e94\u7684\u4e24\u4e2a CTC \u6982\u7387\u5206\u5e03<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5e0f2e37316aef097e61c03c590b6bc8.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5e0f2e37316aef097e61c03c590b6bc8.png\"><\/a>\u548c<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-475280e789fed66d793f7e3864adcc54.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-475280e789fed66d793f7e3864adcc54.png\"><\/a>\uff0c\u9664\u4e86\u8ba1\u7b97\u4e24\u4e2a CTC loss <a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-983e903467f7b756601f327ab42c5516.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-983e903467f7b756601f327ab42c5516.png\"><\/a>\u548c<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7fcdbc3619bb1f9b9ebcab90e32c3c67.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-7fcdbc3619bb1f9b9ebcab90e32c3c67.png\"><\/a>\uff0c\u8fd8\u5f15\u5165 consistency regularization loss \u6765\u7ea6\u675f\u4e24\u4e2a\u5206\u5e03\u7684\u4e00\u81f4\u6027\uff1a<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-05156940133326f37228d8ddcb6d31b9.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-05156940133326f37228d8ddcb6d31b9.png\"><\/a>\u3002\u7cfb\u7edf\u603b\u4f53 loss \u4e3a\uff1a<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1b4d08aa27766ec53e30bebe86cb82c5.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1b4d08aa27766ec53e30bebe86cb82c5.png\"><\/a><\/section>\n<section>\u5176\u4e2d \u03b1 \u4e3a\u63a7\u5236\u6b63\u5219\u7684\u8d85\u53c2\u6570\uff0c\u9ed8\u8ba4\u8bbe\u7f6e\u4e3a 0.2\u3002<\/section>\n<section><\/section>\n<section><strong>Different augmented views<\/strong><\/section>\n<section><\/section>\n<section>\u6211\u4eec\u5bf9\u540c\u4e00\u4e2a\u8f93\u5165 x \u7684\u4e24\u4e2a copy \u72ec\u7acb\u5730\u4f7f\u7528 SpecAugment [4] \u6765\u83b7\u5f97\u4e0d\u540c\u7684 augmented views\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-2ca654656f57a81bfb196b0a2e15c2a8.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-2ca654656f57a81bfb196b0a2e15c2a8.png\"><\/a>\u548c<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-34081c9bddb8378e517f205a2a7a9536.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-34081c9bddb8378e517f205a2a7a9536.png\"><\/a>\u3002SpecAugment \u5305\u542b time warping\u3001frequency masking \u548c time masking\u3002\u7531\u4e8e time warping \u4f1a\u663e\u8457\u6539\u53d8\u8f93\u51fa\u7684\u65f6\u95f4\u6233\uff0c\u56e0\u6b64\u6211\u4eec\u5728\u521b\u5efa copy \u524d\u5148\u5e94\u7528 time warping\uff0c\u9632\u6b62\u4e24\u4e2a\u5206\u652f\u7684\u8f93\u51fa\u5206\u5e03\u5728\u65f6\u95f4\u6233\u4e0a\u4e25\u91cd\u4e0d\u5339\u914d\u3002\u63a5\u7740\uff0c\u5206\u522b\u5bf9\u4e24\u4e2a copy \u72ec\u7acb\u5e94\u7528 frequency masking \u548c time masking\uff0c\u5f97\u5230\u4e86\u00a0<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ffba388dd81f5528e75b1e900a898aeb.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ffba388dd81f5528e75b1e900a898aeb.png\"><\/a>\u548c<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-985628d4d582267f09f6ffcf2adbe73b.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-985628d4d582267f09f6ffcf2adbe73b.png\"><\/a>\u3002\u76f8\u8f83\u4e8e\u666e\u901a\u7684 ASR \u7cfb\u7edf\uff0c\u6211\u4eec\u7279\u610f\u4f7f\u7528\u4e86\u66f4\u5927\u7a0b\u5ea6\u7684 time masking\u3002<\/section>\n<section><\/section>\n<section><strong>Consistency regularization loss<\/strong><\/section>\n<section><\/section>\n<section>\u6211\u4eec\u5728CTC \u5206\u5e03\u7684\u6bcf\u4e00\u5e27\u4e0a\u5e94\u7528 consistency regularization\uff0c\u901a\u8fc7\u6700\u5c0f\u5316\u6bcf\u4e00\u5bf9\u5206\u5e03<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-625a843beec7e32b5219a05c1dd1c780.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-625a843beec7e32b5219a05c1dd1c780.png\"><\/a>\u548c<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-51f32e8a02fc96675d5c39772fdb596a.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-51f32e8a02fc96675d5c39772fdb596a.png\"><\/a>\u4e4b\u95f4\u7684\u53cc\u5411 KL \u6563\u5ea6\uff1a<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5a4fe378e8172077d430b9bb04399415.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5a4fe378e8172077d430b9bb04399415.png\"><\/a> \u548c <a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-09c4bf71975346ff51a66e1572325d95.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-09c4bf71975346ff51a66e1572325d95.png\"><\/a>\u3002\u6b64\u5904\uff0csg \u8868\u793a stop-gradient\uff0c\u9632\u6b62\u8fd9\u4e00\u9879\u7684\u68af\u5ea6\u5f71\u54cd\u76ee\u6807\u5206\u5e03\u3002Consistency regularization loss \u516c\u5f0f\u4e3a\uff1a<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-a540186363d1a27432d562d918882fed.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-a540186363d1a27432d562d918882fed.png\"><\/a><\/section>\n<section><strong>\u65b9\u6cd5\u89e3\u91ca<\/strong><\/section>\n<section><\/section>\n<section>\u8bba\u6587\u4ece\u4e09\u4e2a\u4e0d\u540c\u7684\u89d2\u5ea6\u6765\u89e3\u91ca CR-CTC \u7684\u672c\u8d28\u884c\u4e3a\uff1a1\uff09self-distillation\uff1b2\uff09masked prediction\uff1b3\uff09peak suppression\u3002<\/section>\n<section><\/section>\n<section><strong>Self-distillation<\/strong><\/section>\n<section><\/section>\n<section>\u5f53\u6211\u4eec\u5728\u8bad\u7ec3\u4e2d\u4f7f\u7528 dropout [5] \u548c stochastic depth [6] \u7b49\u6a21\u578b\u6b63\u5219\u6280\u672f\uff0c\u53ef\u4ee5\u770b\u4f5c\u6211\u4eec\u6b63\u5728\u9690\u5f0f\u5730\u8bad\u7ec3\u968f\u673a\u91c7\u6837\u7684\u4e0d\u540c sub-model\uff0c\u8fd9\u4e9b sub-model \u6700\u7ec8\u88ab\u96c6\u6210\u4e3a\u4e00\u4e2a ensemble \u7528\u4e8e\u63a8\u7406\u3002\u4e0e R-Drop [7] \u548c cosub [8] \u7c7b\u4f3c\uff0cCR-CTC \u5728\u8fdb\u884c\u5bf9\u4e0d\u540c sub-model \u4e4b\u95f4\u7684 self-distillation\uff0c\u76d1\u7763\u4fe1\u53f7\u4e3a\u5bf9\u65b9\u63d0\u4f9b\u7684\u5e27\u7ea7\u522b\u7684 token \u5206\u5e03\u3002\u53e6\u5916\uff0cCR-CTC \u4f7f\u7528\u4e86\u4e0d\u540c\u7684 augmented views\uff08\u4ee5\u53ca\u66f4\u5927\u7a0b\u5ea6\u7684 time-masking\uff09\uff0c\u8ba9\u8fd9\u4e9b sub-model \u63a5\u89e6\u8f93\u5165\u6570\u636e\u7684\u4e0d\u540c\u89d2\u5ea6\u7684\u4fe1\u606f\uff0c\u52a0\u5f3a\u4ed6\u4eec\u9884\u6d4b\u7684\u591a\u6837\u6027\uff0c\u8fd9\u6837\u6709\u5229\u4e8e\u66f4\u4e30\u5bcc\u3001\u66f4\u5b8c\u5907\u7684\u77e5\u8bc6\u84b8\u998f\u3002<\/section>\n<section><\/section>\n<section><strong>Masked prediction<\/strong><\/section>\n<section><\/section>\n<section>\u5728 CR-CTC \u4e2d\uff0c\u90a3\u4e9b\u8986\u76d6\u5728 time masking \u533a\u57df\u7684\u5e27\uff0c\u88ab\u8981\u6c42\u7740\u57fa\u4e8e\u5176\u4ed6\u6ca1\u6709\u88ab masked \u7684\u533a\u57df\uff0c\u53bb\u9884\u6d4b\u5bf9\u65b9\u63d0\u4f9b\u7684 token \u5206\u5e03\u3002\u8fd9\u4e2a\u8fc7\u7a0b\u7c7b\u4f3c\u4e8e masked-based \u81ea\u76d1\u7763\u6a21\u578b [9,10,11]\uff0c\u9f13\u52b1\u6a21\u578b\u53bb\u5b66\u4e60\u975e mask \u90e8\u5206\u7684\u4e0a\u4e0b\u6587\u8868\u5f81\u4fe1\u606f\uff0c\u5e76\u53d1\u6398\u6a21\u578b\u9690\u5f0f\u7684\u8bed\u8a00\u5efa\u6a21\u80fd\u529b\u3002\u6211\u4eec\u5728 CR-CTC \u4e2d\u4f7f\u7528\u4e0d\u540c\u7684 augmented views\uff0c\u51cf\u5c11\u4e24\u8fb9\u540c\u65f6\u88ab\u8986\u76d6\u5728 time masking \u533a\u57df\u7684\u5e27\u7684\u51fa\u73b0\uff0c\u63d0\u9ad8\u8fd9\u4e9b\u88ab masked \u4f4d\u7f6e\u6240\u63a5\u6536\u7684 token \u5206\u5e03\u7684\u8d28\u91cf\u3002\u53e6\u5916\uff0c\u4f7f\u7528\u66f4\u5927\u7a0b\u5ea6\u7684 time masking \u53ef\u4ee5\u52a0\u5f3a masked prediction \u884c\u4e3a\uff0c\u8fdb\u800c\u589e\u5f3a\u6a21\u578b\u5bf9\u4e0a\u4e0b\u6587\u8868\u5f81\u4fe1\u606f\u7684\u5b66\u4e60\u3002\u00a0<\/section>\n<section><\/section>\n<section><strong>Peak suppression<\/strong><\/section>\n<section><\/section>\n<section>\u4f17\u6240\u5468\u77e5\uff0cCTC \u901a\u5e38\u4f1a\u5b66\u4e60\u5230\u975e\u5e38\u5c16\u7684\u6982\u7387\u5206\u5e03\u3002\u5982 Figure 2 (left) \u6240\u793a\uff0cnon-blank token \u53ea\u5360 1 \u5e27\uff0c\u5176\u4ed6\u7684\u90fd\u662f blank\uff0c\u5b83\u4eec\u7684\u6982\u7387\u90fd\u975e\u5e38\u9ad8\u3002\u8fd9\u79cd\u73b0\u8c61\u8868\u660e\u6a21\u578b\u6709\u53ef\u80fd\u5df2\u7ecf\u8fc7\u62df\u5408\u4e86\uff0c\u6cdb\u5316\u80fd\u529b\u4e0d\u5f3a\u3002CR-CTC \u7684 consistency regularization \u5f15\u5bfc\u7740\u6a21\u578b\u5b66\u4e60\u4e24\u8fb9\u5206\u5e03\u7684\u5e73\u5747\uff0c\u8fd9\u4f7f\u5f97\u6a21\u578b\u5b66\u4e60\u5230\u7684 CTC \u5206\u5e03\u4f1a\u66f4\u52a0\u5e73\u6ed1\u3002\u8fd9\u4e2a peak suppression \u884c\u4e3a\u51cf\u5c11\u4e86\u5728\u8bad\u7ec3\u6570\u636e\u4e0a\u7684\u8fc7\u5ea6\u7f6e\u4fe1\uff0c\u4ece\u800c\u589e\u5f3a\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u3002\u5982 Figure 2 (right) \u6240\u793a\uff0cCR-CTC \u5b66\u4e60\u5230\u7684\u5206\u5e03\u66f4\u52a0\u5e73\u6ed1\uff0c\u6982\u7387\u66f4\u4f4e\uff0c\u4f34\u968f\u7740\u66f4\u591a non-blank \u7684 repeat \u51fa\u73b0\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-303c7048dc39b94f3ea4240f2c8a633d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-303c7048dc39b94f3ea4240f2c8a633d.png\"><\/a><\/section>\n<section><strong>\u5b9e\u9a8c\u7ed3\u679c<\/strong><\/section>\n<section><\/section>\n<section>\u8bba\u6587\u4e3b\u8981\u4f7f\u7528 Zipformer [12] \u4f5c\u4e3a speech encoder \u8fdb\u884c\u5b9e\u9a8c\u9a8c\u8bc1\u3002\u7531\u4e8e CR-CTC \u8bad\u7ec3\u65f6\u9700\u8981\u8fdb\u884c\u4e24\u6b21 forward\uff0c\u6211\u4eec\u5bf9 CR-CTC \u6a21\u578b\u7684 batch size \u548c epoch \u6570\u90fd\u8bbe\u7f6e\u4e3a\u6807\u51c6 CTC \u6a21\u578b\u7684\u4e00\u534a\uff0c\u6765\u786e\u4fdd\u4e24\u8005\u8bad\u7ec3\u4ee3\u4ef7\u53ef\u6bd4\u8f83\u3002\u5177\u4f53\u4f7f\u7528\u7684 GPU \u6570\u91cf\u548c epoch \u6570\u5728\u8bba\u6587\u9644\u5f55\u4e2d\u3002<\/section>\n<section><\/section>\n<section><strong>\u4e0e SOTA \u6a21\u578b\u76f8\u6bd4\u8f83<\/strong><\/section>\n<section><\/section>\n<section>Table 1\u30012\u30013 \u5206\u522b\u5c55\u793a\u4e86\u4e0d\u540c\u6a21\u578b\u5728 LibriSpeech\u3001Aishell-1\u3001GigaSpeech \u4e09\u4e2a\u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0\uff08\u4e0d\u4f9d\u8d56\u5916\u90e8\u8bad\u7ec3\u6570\u636e\u548c\u5916\u90e8\u8bed\u8a00\u6a21\u578b\uff09\u3002\u603b\u7684\u6765\u8bf4\uff0cCR-CTC \u7684\u6027\u80fd\u663e\u8457\u8d85\u8d8a\u6807\u51c6 CTC\uff0c\u548c CTC\/AED \u4e0e Transducer \u6a21\u578b\u6548\u679c\u76f8\u5f53\u3002\u53e6\u5916\uff0c\u6302\u4e0a CR-CTC \u8054\u5408\u8bad\u7ec3\uff0c\u53ef\u4ee5\u8fdb\u4e00\u6b65\u63d0\u5347 CTC\/AED \u548c Transducer \u7684\u6027\u80fd\u3002\u5728\u8fd9\u4e09\u4e2a\u6570\u636e\u96c6\u4e0a\uff0c\u6211\u4eec\u53d6\u5f97\u4e86\u65b0\u7684 SOTA \u7ed3\u679c\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c7e0b76e8967384acef8158f8dae7a02.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c7e0b76e8967384acef8158f8dae7a02.png\"><\/a><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-228f2c51beceb749ec275666e4d2acda.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-228f2c51beceb749ec275666e4d2acda.png\"><\/a><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-27c7495b04fae7a32525d6b585efccb1.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-27c7495b04fae7a32525d6b585efccb1.png\"><\/a><\/section>\n<section><strong>\u6d88\u878d\u5b9e\u9a8c<\/strong><\/section>\n<section><\/section>\n<section>Table 4\u30015\u30016 \u5206\u522b\u5c55\u793a\u4e86 CR-CTC \u5173\u4e8e\u4e0d\u540c\u89e3\u91ca\u89d2\u5ea6 self-distillation\u3001masked prediction\u3001peak suppression \u7684\u6d88\u878d\u5b9e\u9a8c\u7ed3\u679c\uff0c\u5177\u4f53\u8bf4\u660e\u53ef\u53c2\u8003\u8bba\u6587\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-18d5c2cf6e1411230a0262aeac8aaf37.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-18d5c2cf6e1411230a0262aeac8aaf37.png\"><\/a><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-d220176a5f9e11b516f4c744ac81884d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-d220176a5f9e11b516f4c744ac81884d.png\"><\/a><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5332cec0c2a6e6b880e38695fbdf42be.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5332cec0c2a6e6b880e38695fbdf42be.png\"><\/a><\/section>\n<section><strong>\u4e0e\u6302\u4e00\u4e2a auxiliary head \u8054\u5408\u8bad\u7ec3\u76f8\u6bd4\u8f83<\/strong><\/section>\n<section><\/section>\n<section>\u60f3\u8981\u63d0\u5347 CTC \u7cfb\u7edf\u7684\u6027\u80fd\uff0c\u4e00\u4e2a\u6700\u76f4\u63a5\u7684\u65b9\u6cd5\u4fbf\u662f\u6302\u4e00\u4e2a AED head \u6216\u8005\u4e00\u4e2a Transducer head \u8054\u5408\u8bad\u7ec3\u3002\u5982 Table 7 \u6240\u793a\uff0cCR-CTC \u7684\u6027\u80fd\u660e\u663e\u8d85\u8fc7\u8fd9\u4e24\u4e2a\u65b9\u6cd5\uff0c\u53c2\u6570\u8fd8\u66f4\u5c11\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-39dc8f3d94ee78b209bd08e386611510.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-39dc8f3d94ee78b209bd08e386611510.png\"><\/a><\/section>\n<section><strong>\u5728 Conformer \u6a21\u578b\u4e0a\u9a8c\u8bc1<\/strong><\/section>\n<section><\/section>\n<section>\u5982 Table 17 \u6240\u793a\uff0c\u4f7f\u7528 Conformer [13] \u4f5c\u4e3a speech encoder \u65f6\uff0cCR-CTC \u540c\u6837\u53ef\u4ee5\u663e\u8457\u63d0\u5347 CTC \u7684\u6027\u80fd\uff0c\u5e76\u4e14\u7565\u5fae\u8d85\u8fc7 CTC\/AED \u548c Transducer\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f1fbed0fbe2b982803fbfe4e6b35bf7f.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f1fbed0fbe2b982803fbfe4e6b35bf7f.png\"><\/a><\/section>\n<section><strong><sup>\u53c2\u8003<\/sup><\/strong><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[1] Graves, A., Fern\u00e1ndez, S., Gomez, F., &amp; Schmidhuber, J. (2006, June). Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks. In Proceedings of the 23rd international conference on Machine learning (pp. 369-376).<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[2] Graves, A. (2012). Sequence transduction with recurrent neural networks. arXiv preprint arXiv:1211.3711.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[3] Watanabe, S., Hori, T., Kim, S., Hershey, J. R., &amp; Hayashi, T. (2017). Hybrid CTC\/attention architecture for end-to-end speech recognition. IEEE Journal of Selected Topics in Signal Processing, 11 (8), 1240-1253.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[4] Park, D. S., Chan, W., Zhang, Y., Chiu, C. C., Zoph, B., Cubuk, E. D., &amp; Le, Q. V. (2019). Specaugment: A simple data augmentation method for automatic speech recognition. arXiv preprint arXiv:1904.08779.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[5] Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., &amp; Salakhutdinov, R. (2014). Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research, 15 (1), 1929-1958.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[6] Huang, G., Sun, Y., Liu, Z., Sedra, D., &amp; Weinberger, K. Q. (2016). Deep networks with stochastic depth. In Computer Vision\u2013ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11\u201314, 2016, Proceedings, Part IV 14 (pp. 646-661). Springer International Publishing.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[7] Wu, L., Li, J., Wang, Y., Meng, Q., Qin, T., Chen, W., &#8230; &amp; Liu, T. Y. (2021). R-drop: Regularized dropout for neural networks. Advances in Neural Information Processing Systems, 34, 10890-10905.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[8] Touvron, H., Cord, M., Oquab, M., Bojanowski, P., Verbeek, J., &amp; J\u00e9gou, H. (2023). Co-training 2L submodels for visual recognition. In Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition (pp. 11701-11710).<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[9] Devlin, J. (2018). Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[10] Baevski, A., Zhou, Y., Mohamed, A., &amp; Auli, M. (2020). wav2vec 2.0: A framework for self-supervised learning of speech representations. Advances in neural information processing systems, 33, 12449-12460.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[11] Hsu, W. N., Bolte, B., Tsai, Y. H. H., Lakhotia, K., Salakhutdinov, R., &amp; Mohamed, A. (2021). Hubert: Self-supervised speech representation learning by masked prediction of hidden units. IEEE\/ACM transactions on audio, speech, and language processing, 29, 3451-3460.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[12] Yao, Z., Guo, L., Yang, X., Kang, W., Kuang, F., Yang, Y., &#8230; &amp; Povey, D. (2023, October). Zipformer: A faster and better encoder for automatic speech recognition. In The Twelfth International Conference on Learning Representations.<\/sup><\/section>\n<section><sup><br \/><\/sup><\/section>\n<section><sup>[13] Gulati, A., Qin, J., Chiu, C. C., Parmar, N., Zhang, Y., Yu, J., &#8230; &amp; Pang, R. (2020). Conformer: Convolution-augmented transformer for speech recognition. arXiv preprint arXiv:2005.08100.<\/sup><\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-07-5\" target=\"_blank\">ICLR 2025\uff5c\u5c0f\u7c73\u65b0\u4e00\u4ee3Kaldi\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5CR-CTC\uff0c\u7eafCTC\u6027\u80fd\u5b9e\u73b0SOTA<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:ICLR 2025\uff5c\u5c0f [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","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":"","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":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","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":[27],"tags":[],"class_list":["post-38134","post","type-post","status-publish","format-standard","hentry","category-news"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>ICLR 2025\uff5c\u5c0f\u7c73\u65b0\u4e00\u4ee3Kaldi\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5CR-CTC\uff0c\u7eafCTC\u6027\u80fd\u5b9e\u73b0SOTA - \u4e00\u8d77AI\u6280\u672f<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/17aitech.com\/?p=38134\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=38134\",\"url\":\"https:\/\/17aitech.com\/?p=38134\",\"name\":\"ICLR 2025\uff5c\u5c0f\u7c73\u65b0\u4e00\u4ee3Kaldi\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5CR-CTC\uff0c\u7eafCTC\u6027\u80fd\u5b9e\u73b0SOTA - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=38134#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=38134#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1daf4275c3cb8450850a3f96506b97a2.png\",\"datePublished\":\"2025-02-10T13:01:39+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=38134#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=38134\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=38134#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1daf4275c3cb8450850a3f96506b97a2.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1daf4275c3cb8450850a3f96506b97a2.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=38134#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"ICLR 2025\uff5c\u5c0f\u7c73\u65b0\u4e00\u4ee3Kaldi\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5CR-CTC\uff0c\u7eafCTC\u6027\u80fd\u5b9e\u73b0SOTA\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/17aitech.com\/#website\",\"url\":\"https:\/\/17aitech.com\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/17aitech.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/image\/\",\"url\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"contentUrl\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\/\/17aitech.com\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"ICLR 2025\uff5c\u5c0f\u7c73\u65b0\u4e00\u4ee3Kaldi\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5CR-CTC\uff0c\u7eafCTC\u6027\u80fd\u5b9e\u73b0SOTA - \u4e00\u8d77AI\u6280\u672f","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/17aitech.com\/?p=38134","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=38134","url":"https:\/\/17aitech.com\/?p=38134","name":"ICLR 2025\uff5c\u5c0f\u7c73\u65b0\u4e00\u4ee3Kaldi\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5CR-CTC\uff0c\u7eafCTC\u6027\u80fd\u5b9e\u73b0SOTA - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=38134#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=38134#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1daf4275c3cb8450850a3f96506b97a2.png","datePublished":"2025-02-10T13:01:39+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=38134#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=38134"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=38134#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1daf4275c3cb8450850a3f96506b97a2.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-1daf4275c3cb8450850a3f96506b97a2.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=38134#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"ICLR 2025\uff5c\u5c0f\u7c73\u65b0\u4e00\u4ee3Kaldi\u8bed\u97f3\u8bc6\u522b\u7b97\u6cd5CR-CTC\uff0c\u7eafCTC\u6027\u80fd\u5b9e\u73b0SOTA"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/#\/schema\/person\/image\/","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38134","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=38134"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38134\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=38134"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=38134"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=38134"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}