{"id":9986,"date":"2026-07-08T16:46:44","date_gmt":"2026-07-08T08:46:44","guid":{"rendered":"https:\/\/infernews.com\/blog\/official-repository-of-the-ieee-ojsp-paper-quot-speaker-disentangled-chunk-wise\/"},"modified":"2026-07-08T16:46:44","modified_gmt":"2026-07-08T08:46:44","slug":"official-repository-of-the-ieee-ojsp-paper-quot-speaker-disentangled-chunk-wise","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/official-repository-of-the-ieee-ojsp-paper-quot-speaker-disentangled-chunk-wise\/","title":{"rendered":"speaker_disentangled_hubert\uff1a\u7528 HuBERT \u91cd\u65b0\u505a\u597d\u97f3\u7bc0\u5207\u5206"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/07\/pasted-94e3b4742ea4.jpg\" alt=\"Repository image for ryota-komatsu\/speaker_disentangled_hubert\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u505a\u8a9e\u97f3 tokenization\uff0c\u6700\u9ebb\u7169\u7684\u60c5\u6cc1\u5f80\u5f80\u4e0d\u662f\u97f3\u7bc0\u908a\u754c\u627e\u4e0d\u5230\uff0c\u800c\u662f\u6a21\u578b\u628a\u8b1b\u8005\u7279\u5fb5\u4e00\u4f75\u5b78\u9032\u53bb\uff0c\u6700\u5f8c\u5206\u51fa\u4f86\u7684 token \u6df7\u96dc\u4e86\u8072\u7dda\u800c\u4e0d\u662f\u8a9e\u8a00\u5167\u5bb9\u3002speaker_disentangled_hubert \u5c6c\u65bc<strong>\u8a9e\u97f3\u6a21\u578b\u9805\u76ee<\/strong>\uff0c\u6838\u5fc3\u5de5\u4f5c\u662f\u628a\u539f\u59cb\u8a9e\u97f3\u6574\u7406\u6210\u8f03\u7d14\u6de8\u7684 syllabic token\uff0c\u8b93\u5f8c\u7e8c\u7684 speech language model \u66f4\u5bb9\u6613\u5b78\u5230\u53e5\u6cd5\u8207\u8a9e\u610f\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u6709\u505a\u6cd5\u591a\u6578\u6cbf\u7528 pretrained HuBERT \u7684 teacher-student distillation\uff0c\u518d\u7528 utterance-level cross-entropy \u76ee\u6a19\u53bb\u7d44\u7e54 frame \u8868\u5fb5\uff1b\u4f5c\u8005\u8a8d\u70ba\u9019\u7a2e fixed \u7bc4\u5f0f\u6703\u4ee4\u5b78\u751f\u6a21\u578b\u504f\u5411\u9810\u6e2c speaker identity\u3002\u9019\u500b\u9805\u76ee\u6539\u7528 <strong>Speaker-Disentangled Chunk-Wise Regression for Syllabic Tokenization<\/strong>\uff0c\u5728 fixed-length chunks \u5167\uff0c\u5c07\u52a0\u5165 speaker \u64fe\u52d5\u7684 student representations \u62c9\u56de\u4e7e\u6de8\u7684 teacher targets\uff0c\u91cd\u9ede\u4e0d\u662f\u518d\u52a0\u66f4\u591a\u6a19\u7c64\uff0c\u800c\u662f\u76f4\u63a5\u524a\u5f31\u8b1b\u8005\u8cc7\u8a0a\u5c0d\u97f3\u7bc0\u5207\u5206\u7684\u5e72\u64fe\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5f9e\u7d50\u679c\u4f86\u770b\uff0c\u5b83\u7684\u53d6\u5411\u76f8\u7576\u660e\u78ba\uff1a\u5be7\u9858\u628a\u8a13\u7df4\u8a2d\u8a08\u505a\u5f97\u66f4\u7d30\uff0c\u4e5f\u8981\u63db\u53d6\u66f4\u4e7e\u6de8\u7684 syllabic segments\u3002\u8ad6\u6587\u6307\u51fa\uff0c\u9019\u5957\u65b9\u6cd5\u5728 syllable boundary detection \u8207 syllabic segment clustering \u90fd\u505a\u5230 state-of-the-art\uff1b\u518d\u5f80\u4e0b\u63a5 speech language model\uff0c\u57fa\u65bc\u9019\u6279 syllabic tokens \u7684\u6a21\u578b\uff0c\u5c0d\u6bd4 phone-level SpiRit-LM\uff0c\u5728 syntactic and semantic understanding \u53d6\u5f97 7% relative improvement\u3002\u5c0d\u7814\u7a76\u8a9e\u97f3\u8868\u793a\u5b78\u7fd2\u3001\u4f4e\u8cc7\u6e90\u8a9e\u97f3\u5efa\u6a21\uff0c\u6216\u8005\u60f3\u628a\u96e2\u6563\u8a9e\u97f3\u55ae\u4f4d\u63a5\u5230\u8a9e\u8a00\u6a21\u578b\u7684\u4eba\uff0c\u9019\u500b\u65b9\u5411\u5e7e\u6709\u53c3\u8003\u50f9\u503c\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u91dd\u5c0d HuBERT teacher-student distillation \u5bb9\u6613\u6df7\u5165 speaker identity \u7684\u554f\u984c<\/li>\n<li>\u7528 chunk-wise regression \u53d6\u4ee3\u55ae\u7d14 utterance-level cross-entropy \u8a13\u7df4\u908f\u8f2f<\/li>\n<li>\u4efb\u52d9\u7126\u9ede\u5305\u62ec syllable boundary detection \u8207 syllabic segment clustering<\/li>\n<li>\u76f8\u95dc\u8108\u7d61\u6db5\u84cb pretrained HuBERT\u3001speech language model\uff0c\u4ee5\u53ca phone-level SpiRit-LM<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u6e2c\u8a66\u8207\u7406\u89e3\u65b9\u5f0f\u4ea6\u7b97\u76f4\u63a5\uff1a\u9805\u76ee\u63d0\u4f9b Paper\u3001Code\u3001Model \u8207\u7db2\u9801\u4e0a\u7684 resynthesis samples\u3001syllable segmentation \u793a\u4f8b\uff0c\u5148\u807d\u91cd\u5efa\u8a9e\u97f3\uff0c\u518d\u770b\u97f3\u7bc0\u5206\u6bb5\uff0c\u5df2\u7d93\u53ef\u4ee5\u5224\u65b7 tokenization \u6709\u6c92\u6709\u4fdd\u4f4f\u8a9e\u8a00\u7d50\u69cb\u3002\u73fe\u6709\u8cc7\u8a0a\u672a\u5b8c\u6574\u5217\u51fa\u90e8\u7f72\u6d41\u7a0b\u8207\u5b89\u88dd\u7d30\u7bc0\uff0c\u6240\u4ee5\u8f03\u9069\u5408\u4f5c\u70ba\u7814\u7a76\u578b\u9805\u76ee\u95b1\u8b80\u3001\u91cd\u73fe\u8207\u6bd4\u8f03\uff1b\u4f7f\u7528\u6642\u4e5f\u8981\u7559\u610f\uff0c\u5b83\u7684\u9a57\u8b49\u4e3b\u8981\u5efa\u57fa\u65bc LibriSpeech\uff0c\u8de8\u8a9e\u8005\u4ee5\u5916\u7684\u8a9e\u7a2e\u3001\u9304\u97f3\u689d\u4ef6\u8207\u8cc7\u6599\u5206\u4f48\uff0c\u4ecd\u7136\u503c\u5f97\u518d\u505a\u984d\u5916\u78ba\u8a8d\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/ryota-komatsu.github.io\/speaker_disentangled_hubert\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/ryota-komatsu\/speaker_disentangled_hubert\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/arxiv.org\/pdf\/2607.04064\" rel=\"noopener noreferrer\"><strong>Paper<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u8a9e\u97f3\u5207\u6210\u97f3\u7bc0\uff0c\u96e3\u8655\u5f80\u5f80\u5514\u4fc2\u5206\u6bb5\u672c\u8eab\uff0c\u800c\u4fc2\u6a21\u578b\u597d\u6613\u8a18\u4f4f\u8b1b\u8005\u8072\u7dda\u3002speaker_disentangled_hubert \u91dd\u5c0d\u5462\u500b\u504f\u5dee\u843d\u624b\uff0c\u4ee4 token \u66f4\u63a5\u8fd1\u8a9e\u8a00\u5167\u5bb9\u3002<\/p>\n","protected":false},"author":8,"featured_media":9985,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,128,199],"tags":[],"class_list":["post-9986","post","type-post","status-publish","format-standard","hentry","category-133","category-128","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9986","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=9986"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9986\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9985"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9986"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9986"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9986"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}