{"id":4977,"date":"2026-03-19T14:50:54","date_gmt":"2026-03-19T06:50:54","guid":{"rendered":"https:\/\/www.15zhi.net\/blog\/?p=4977"},"modified":"2026-03-19T14:50:54","modified_gmt":"2026-03-19T06:50:54","slug":"202603%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-estimate-a-real-time-speech-transmission-index-estimator-with-speech-enhancement-auxiliary-task-using-self-attention-feature-pyramid-network","status":"publish","type":"post","link":"https:\/\/www.15zhi.net\/blog\/202603%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-estimate-a-real-time-speech-transmission-index-estimator-with-speech-enhancement-auxiliary-task-using-self-attention-feature-pyramid-network\/","title":{"rendered":"202603\u8bba\u6587\u7814\u8bfb-eSTImate: A Real-time Speech Transmission Index Estimator With Speech Enhancement Auxiliary Task Using Self-Attention Feature Pyramid Network"},"content":{"rendered":"\n

2026\u5e743\u670819\u65e5-yuehan<\/p>\n\n\n\n

\u4f5c\u8005\uff1aBajian Xiang, Hongkun Liu, Zedong Wu, Su Shen, Xiangdong Zhang<\/p>\n\n\n\n

\u6765\u6e90\uff1ainterspeech 2023<\/p>\n\n\n\n

\u5355\u4f4d\uff1aYousonic Technology, Beijing, China<\/p>\n\n\n\n

1\u3001\u80cc\u666f<\/h2>\n\n\n\n

STI \u7684\u6838\u5fc3\u4ef7\u503c\u4e0e\u5b9a\u4e49\uff1aSTI \u662f\u8bc4\u4f30\u8bed\u97f3\u53ef\u61c2\u5ea6\u7684\u5173\u952e\u5ba2\u89c2\u6307\u6807\uff0c\u53d6\u503c\u8303\u56f40-1\uff08\u503c\u8d8a\u9ad8\u53ef\u61c2\u5ea6\u8d8a\u597d\uff09\uff0c\u5e76\u6309 IEC 60268-16:2020 \u6807\u51c6\u5206\u4e3a 5 \u4e2a\u7b49\u7ea7\uff1abad\uff08<0.30\uff09\u3001poor\uff080.30-0.45\uff09\u3001fair\uff080.45-0.60\uff09\u3001good\uff080.60-0.75\uff09\u3001excellent\uff08>0.75\uff09\uff0c\u4e0e\u8bed\u97f3\u53ef\u61c2\u5ea6\u5f3a\u76f8\u5173\u3002<\/p>\n\n\n\n

\u4f20\u7edf STI \u6d4b\u91cf\u65b9\u6cd5\u7684\u7f3a\u9677\uff1a<\/p>\n\n\n\n

\u76f4\u63a5\u6cd5\uff1a\u91c7\u7528 7 \u4e2a\u500d\u9891\u5e26 + 14 \u4e2a\u8c03\u5236\u9891\u7387\u751f\u6210\u6d4b\u8bd5\u4fe1\u53f7\uff0c\u624b\u6301\u8bbe\u5907\u6d4b\u91cf\u9700 15 \u5206\u949f\uff0c\u65e0\u6cd5\u5b9e\u65f6\u5e94\u7528<\/p>\n\n\n\n

\u95f4\u63a5\u6cd5\uff1a\u4ece\u623f\u95f4\u51b2\u6fc0\u54cd\u5e94\uff08RIR\uff09\u63a8\u5bfc STI\uff0c\u4f46\u5b9e\u9645\u573a\u666f\u4e2d\u5b9e\u65f6\u83b7\u53d6 RIR \u96be\u5ea6\u6781\u5927<\/p>\n\n\n\n

\u73b0\u6709\u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u7684\u4e0d\u8db3\uff1a\u8986\u76d6\u8303\u56f4\u6709\u9650\uff1a\u5982 CNN \u6a21\u578b\u4ec5\u80fd\u9884\u6d4b STI\u22650.60 \u7684\u573a\u666f\uff0c\u65e0\u6cd5\u533a\u5206 bad\/poor \u7b49\u7ea7\uff08STI<0.30\uff09\uff1b \u7cbe\u5ea6\u4e0e\u8303\u56f4\u96be\u4ee5\u517c\u987e\uff1aTAE-CNN \u62d3\u5bbd\u4e86\u6d4b\u91cf\u8303\u56f4\uff0c\u4f46\u7cbe\u5ea6\u4e0b\u964d\uff1b \u591a\u53c2\u6570\u8054\u5408\u4f30\u8ba1\u7684\u5f0a\u7aef\uff1a\u5c06 STI \u4e0e\u5176\u4ed6\u623f\u95f4\u58f0\u5b66\u53c2\u6570\u8054\u5408\u9884\u6d4b\u65f6\uff0c\u56e0 STI \u53d7\u566a\u58f0\u5f71\u54cd\u3001\u65f6\u95f4\u6ce2\u52a8\u6027\u8fdc\u5927\u4e8e\u5176\u4ed6\u53c2\u6570\uff0c\u4f1a\u964d\u4f4e\u5176\u5b9e\u65f6\u9884\u6d4b\u6027\u80fd\uff1b \u5bf9\u4f4e STI \u503c\uff08<0.30\uff09\u548c\u9ad8 STI \u503c\uff08>0.80\uff09\u7684\u4f30\u8ba1\u7cbe\u5ea6\u4e0d\u8db3\u3002<\/p>\n\n\n\n

2\u3001\u4e3b\u8981\u8d21\u732e<\/h2>\n\n\n\n

\u63d0\u51fa\u4e00\u6b3e\u76f2 STI \u4f30\u8ba1\u6a21\u578b\uff0c\u65e0\u9700 RIR\u3001\u5e72\u51c0\u8bed\u97f3\u3001\u566a\u58f0\u7b49\u8f85\u52a9\u4fe1\u606f\uff0c\u4ec5\u4ece\u89c2\u6d4b\u8bed\u97f3\u4fe1\u53f7\u76f4\u63a5\u63a8\u5bfc STI\uff0c\u540c\u65f6\u6ee1\u8db3\uff1a\u5168 STI \u8303\u56f4\u8986\u76d6\uff080-1\uff09\u3001\u9ad8\u7cbe\u5ea6\u3001\u5b9e\u65f6\u6027\u3001\u5f3a\u6cdb\u5316\u6027\u3002<\/p>\n\n\n\n

3\u3001\u65b9\u6cd5<\/h2>\n\n\n\n

eSTImate \u662f\u7aef\u5230\u7aef\u6df1\u5ea6\u5b66\u4e60\u7cfb\u7edf\uff0c\u91c7\u7528\u591a\u4efb\u52a1\u8f85\u52a9\u5b66\u4e60\u6846\u67b6\uff0c\u5c06 STI \u76f2\u4f30\u8ba1\u4f5c\u4e3a\u4e3b\u4efb\u52a1\uff0c\u8bed\u97f3\u589e\u5f3a\uff08SE\uff09\u4f5c\u4e3a\u8f85\u52a9\u4efb\u52a1\uff0c\u7ed3\u5408\u7279\u5f81\u91d1\u5b57\u5854\u7f51\u7edc\uff08FPN\uff09\u548c\u591a\u5934\u81ea\u6ce8\u610f\u529b\u673a\u5236\uff0c\u6a21\u578b\u8bad\u7ec3\u65f6\u5171\u4eab\u7f16\u7801\u5668\uff0c\u63a8\u7406\u65f6\u53ef\u820d\u5f03\u8f85\u52a9\u4efb\u52a1\u89e3\u7801\u5668\u4ee5\u63d0\u5347\u901f\u5ea6\u3002<\/p>\n\n\n\n

\u6a21\u578b\u6574\u4f53\u5206\u4e3a\u4e09\u5927\u6838\u5fc3\u6a21\u5757\uff0c\u8bad\u7ec3\u9636\u6bb5\u7ed3\u6784\u89c1\u4e0b\u56fe\uff1a<\/p>\n\n\n\n

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

4\u3001\u5b9e\u9a8c<\/h2>\n\n\n\n

\u6570\u636e\u96c6<\/strong>\uff1a<\/p>\n\n\n\n

\u4e3a\u5b9e\u73b0\u5168 STI \u8303\u56f4\u8986\u76d6\uff0c\u6784\u5efa\u4e86\u5927\u89c4\u6a21\u5408\u6210\u5355\u901a\u9053\u566a\u58f0\u8bed\u97f3\u6570\u636e<\/p>\n\n\n\n

\u6837\u672c\u91cf\uff1a592,928 \u4e2a\uff0c\u91c7\u6837\u7387 16kHz\uff0c\u65f6\u957f 20-30s<\/p>\n\n\n\n

\u6807\u7b7e\uff1a\u6bcf\u4e2a\u6837\u672c\u9644\u5e26\u5b9e\u65f6 STI \u771f\u5b9e\u503c\u548c\u5bf9\u9f50\u7684\u5e72\u51c0\u8bed\u97f3\uff0c\u8986\u76d6 STI 0-1 \u5168\u8303\u56f4<\/p>\n\n\n\n

\u6570\u636e\u751f\u6210\uff1a
\u5e72\u51c0\u8bed\u97f3\uff1aTIMIT \u6570\u636e\u96c6\uff08630 \u4eba\uff0c8 \u79cd\u7f8e\u56fd\u65b9\u8a00\uff0c\u6bcf\u4eba 10 \u53e5\u6807\u51c6\u53e5\uff09
RIR\uff1a108 \u4e2a\uff0832 \u4e2a\u4e2d\u56fd\u529e\u516c\u5ba4\u5b9e\u6d4b + 76 \u4e2a EchoThief \u5e93\uff09\uff0c\u6df7\u54cd\u65f6\u95f4 T60=0.4-1.0s\uff08\u6bcf 0.1s \u95f4\u9694 18 \u4e2a\uff09
\u566a\u58f0\uff1aTUT \u58f0\u5b66\u573a\u666f\u6570\u636e\u96c6\uff08\u8857\u9053\u73af\u5883\uff09+ \u7c89\u7ea2\u566a\u58f0 + \u591a\u4eba\u4ea4\u8c08\u566a\u58f0\uff0c\u4e3a\u6bcf\u4e2a\u6df7\u54cd\u8bed\u97f3\u968f\u673a\u6dfb\u52a0 3 \u7c7b\u566a\u58f0\uff0c\u6bcf\u7c7b\u566a\u58f0\u8bbe 0dB\/6dB\/20dB \u4e09\u79cd\u4fe1\u566a\u6bd4<\/p>\n\n\n\n

\u8bad\u7ec3\u4e0e\u63a8\u7406\u7ec6\u8282<\/strong>\uff1a<\/p>\n\n\n\n

\u8bad\u7ec3\u53c2\u6570\uff1a\u6279\u6b21\u5927\u5c0f 7\uff0c\u521d\u59cb\u5b66\u4e60\u738710\u22125\uff0c\u6bcf 10 \u4e2a epoch \u51cf\u534a
\u5b9e\u65f6\u6027\uff1a\u5355 4s \u5e27\u63a8\u7406\u65f6\u95f4 \u2014\u2014AMD Ryzen5 5600H CPU \u4e3a 0.1s\uff0c24G RTX 3090 GPU \u4e3a 0.047s\uff0c\u6ee1\u8db3\u5b9e\u65f6\u8981\u6c42<\/p>\n\n\n\n

\u6838\u5fc3\u5b9e\u9a8c\u7ed3\u679c<\/strong>\uff1a<\/p>\n\n\n\n

\u5728 STI \u771f\u5b9e\u503c 0.21-0.95 \u7684\u6d4b\u8bd5\u96c6\u4e0a\uff0c\u6a21\u578b\u8fbe\u5230MAE=0.016\uff0cRMSE=0.021\uff0c\u62df\u5408\u6548\u679c\u6781\u5f3a\uff1a
0.30-0.80 \u533a\u95f4\u9884\u6d4b\u503c\u4e0e\u771f\u5b9e\u503c\u51e0\u4e4e\u5b8c\u5168\u91cd\u5408
0.20-0.30 \u533a\u95f4\u8f7b\u5fae\u9ad8\u4f30\uff0c>0.80 \u533a\u95f4\u8f7b\u5fae\u4f4e\u4f30\uff08\u56e0\u6570\u636e\u96c6\u8be5\u533a\u95f4\u6837\u672c\u8f83\u5c11\uff09
STI 5 \u7b49\u7ea7\u5206\u7c7b\u51c6\u786e\u7387 92.75%\uff0c\u8f7b\u5fae\u504f\u5dee\u5bf9\u5b9e\u9645\u5206\u7c7b\u4efb\u52a1\u65e0\u663e\u8457\u5f71\u54cd<\/p>\n\n\n\n

5\u3001\u8bba\u6587\u603b\u7ed3<\/h2>\n\n\n\n

eSTImate \u6a21\u578b\u901a\u8fc7\u7279\u5f81\u91d1\u5b57\u5854\u8f85\u52a9\u5b66\u4e60+\u591a\u5934\u81ea\u6ce8\u610f\u529b\u673a\u5236+\u81ea\u52a8\u52a0\u6743\u591a\u4efb\u52a1\u635f\u5931\uff0c\u5b9e\u73b0\u4e86\u5168 STI \u8303\u56f4\uff080-1\uff09 \u7684\u9ad8\u7cbe\u5ea6\u76f2\u4f30\u8ba1\uff0c\u5728\u81ea\u5efa\u6570\u636e\u96c6\u4e0a\u8fbe\u5230 MAE=0.016\u3001RMSE=0.021\uff0c\u5206\u7c7b\u51c6\u786e\u7387 92.75%\uff0c\u63a8\u7406\u901f\u5ea6\u6ee1\u8db3\u5b9e\u65f6\u8981\u6c42\uff0c\u76f8\u6bd4\u73b0\u6709 SOTA \u65b9\u6cd5\u7cbe\u5ea6\u63d0\u5347 36%\uff0c\u4e14\u8986\u76d6\u66f4\u5e7f\u6cdb\u7684\u5b9e\u9645\u573a\u666f\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

2026\u5e743\u670819\u65e5-yuehan \u4f5c\u8005\uff1aBajian Xiang, Hongkun Liu, Zedong […]<\/p>\n","protected":false},"author":76,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-4977","post","type-post","status-publish","format-standard","hentry","category-events"],"_links":{"self":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/4977","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\/76"}],"replies":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/comments?post=4977"}],"version-history":[{"count":1,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/4977\/revisions"}],"predecessor-version":[{"id":4979,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/4977\/revisions\/4979"}],"wp:attachment":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/media?parent=4977"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/categories?post=4977"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/tags?post=4977"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}