{"id":1069872,"date":"2022-09-15T22:52:00","date_gmt":"2022-09-16T05:52:00","guid":{"rendered":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/?post_type=msr-blog-post&#038;p=1069872"},"modified":"2024-09-25T01:47:48","modified_gmt":"2024-09-25T08:47:48","slug":"text-based-speech-editing","status":"publish","type":"msr-blog-post","link":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/articles\/text-based-speech-editing\/","title":{"rendered":"\u50cf\u7f16\u8f91\u6587\u672c\u4e00\u6837\u7f16\u8f91\u8bed\u97f3\uff0c\u53ef\u80fd\u5417\uff1f"},"content":{"rendered":"\n<p><em>\u4f5c\u8005\uff1a\u8d75\u4e4b\u6e90\u3001\u7f57\u7fc0<\/em><\/p>\n\n\n\n<p>\u7f16\u8005\u6309\uff1a\u5982\u4eca\u5728\u5404\u79cd\u793e\u4ea4\u7f51\u7edc\u5e73\u53f0\u4e0a\u53d1\u5e03\u7684\u89c6\u9891\uff0c\u56e0\u62cd\u6444\u4fbf\u6377\u3001\u53ef\u5b9e\u65f6\u5206\u4eab\u3001\u4e92\u52a8\u4ea4\u6d41\u7b49\u7279\u70b9\u800c\u6df1\u53d7\u5927\u4f17\u559c\u7231\u3002\u89c6\u9891\u6df1\u523b\u5f71\u54cd\u548c\u6539\u53d8\u4e86\u4eba\u4eec\u89c2\u5bdf\u4e16\u754c\u3001\u8bb0\u5f55\u751f\u6d3b\u548c\u8868\u8fbe\u60c5\u611f\u7684\u65b9\u5f0f\u3002\u7136\u800c\uff0c\u73b0\u5728\u5e02\u9762\u4e0a\u8bb8\u591a\u89c6\u9891\u6216\u97f3\u9891\u526a\u8f91\u8f6f\u4ef6\u4e3a\u4e86\u6ee1\u8db3\u7528\u6237\u9700\u6c42\u5c3d\u7ba1\u62e5\u6709\u4e30\u5bcc\u7684\u529f\u80fd\uff0c\u4f46\u64cd\u4f5c\u5374\u5f88\u590d\u6742\uff0c\u5f88\u591a\u7b80\u5355\u7684\u526a\u8f91\u4efb\u52a1\u90fd\u8fd8\u9700\u8981\u5728\u8f6f\u4ef6\u4e2d\u9010\u5e27\u5bf9\u7167\u786e\u5b9a\u526a\u5207\u65f6\u95f4\u70b9\u3002\u5bf9\u4e8e\u4ee5\u8bed\u97f3\u4e3a\u4e3b\u8981\u80cc\u666f\u58f0\u97f3\u7684\u89c6\u9891\uff0c\u5982\u7ebf\u4e0a\u4f1a\u8bae\u5f55\u50cf\u3001\u6f14\u793a\u89c6\u9891\u3001vlog \u7b49\uff0c\u5982\u679c\u6211\u4eec\u80fd\u901a\u8fc7\u7f16\u8f91\u6587\u672c\u7684\u5f62\u5f0f\uff0c\u76f4\u63a5\u7f16\u8f91\u97f3\u89c6\u9891\u4e2d\u7684\u8bed\u97f3\u5185\u5bb9\uff0c\u8ba9\u97f3\u89c6\u9891\u7684\u7f16\u8f91\u81ea\u52a8\u6839\u636e\u6587\u672c\u5b8c\u6210\uff0c\u90a3\u4e48\u5c06\u5927\u5927\u964d\u4f4e\u97f3\u89c6\u9891\u7684\u7f16\u8f91\u96be\u5ea6\uff0c\u63d0\u9ad8\u521b\u4f5c\u8005\u7684\u6548\u7387\u3002\u4e3a\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u7814\u53d1\u4e86\u4e00\u4e2a\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u3002\u672c\u6587\u5c06\u8be6\u7ec6\u4ecb\u7ecd\u8fd9\u4e2a\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u548c\u7814\u7a76\u5458\u4eec\u7814\u53d1\u7684\u8bed\u97f3\u5408\u6210\u53ca\u586b\u5145\u8bcd\u68c0\u6d4b\u6280\u672f\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>\u65e0\u8bba\u662f\u6f14\u793a\u89c6\u9891\u3001\u6559\u5b66\u89c6\u9891\u3001\u4f1a\u8bae\u5f55\u50cf\u8fd8\u662f\u8bb0\u5f55\u751f\u6d3b\u7247\u6bb5\u7684 vlog\uff0c\u5728\u5f88\u591a\u5b9e\u9645\u7684\u5e94\u7528\u573a\u666f\u4e2d\uff0c\u4eba\u4eec\u5e38\u5e38\u9700\u8981\u91cd\u65b0\u5f55\u5236\u8bed\u97f3\uff08\u89c6\u9891\uff09\u6216\u5bf9\u8bed\u97f3\uff08\u89c6\u9891\uff09\u8fdb\u884c\u7f16\u8f91\u3002\u56e0\u4e3a\u62cd\u6444\u7684\u7d20\u6750\u4e2d\u5f80\u5f80\u4f1a\u5b58\u5728\u5927\u91cf\u505c\u987f\u548c\u8131\u53e3\u800c\u51fa\u3001\u8bcd\u4e0d\u8fbe\u610f\u7684\u8bed\u53e5\uff0c\u6216\u8005\u662f\u5197\u4f59\u7684\u5185\u5bb9\u3002\u4f46\u7531\u4e8e\u58f0\u97f3\u7684\u7279\u6027\uff0c\u6211\u4eec\u6ca1\u6709\u529e\u6cd5\u5728\u5f55\u97f3\u5e95\u672c\u7684\u57fa\u7840\u4e0a\u53bb\u4fee\u6539\u5b57\u8bcd\uff0c\u53ea\u80fd\u4e00\u5e27\u4e00\u5e27\u5728\u526a\u8f91\u4e0a\u4e0b\u529f\u592b\uff0c\u56e0\u6b64\u58f0\u97f3\u7684\u526a\u8f91\u5de5\u4f5c\u7e41\u7410\u53c8\u5145\u6ee1\u6311\u6218\u3002\u5982\u679c\u62e5\u6709\u4e00\u4e2a\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\uff0c\u53ef\u4ee5\u901a\u8fc7\u76f4\u63a5\u7f16\u8f91\u8bed\u97f3\u5bf9\u5e94\u7684\u6587\u672c\uff0c\u5b8c\u6210\u5bf9\u8bed\u97f3\uff08\u89c6\u9891\uff09\u7684\u7f16\u8f91\uff0c\u90a3\u4e48\u666e\u901a\u7528\u6237\u4e5f\u80fd\u6210\u4e3a\u4e00\u4e2a\u6709\u521b\u610f\u7684\u526a\u8f91\u5e08\uff0c\u628a\u4e00\u6bb5\u5197\u6742\u7684\u97f3\u89c6\u9891\u53d8\u5f97\u6e05\u6670\u3001\u81ea\u7136\u53c8\u4e13\u4e1a\u3002<\/p>\n\n\n\n<p>\u5e02\u573a\u4e0a\u73b0\u5728\u5df2\u7ecf\u6709\u4e00\u4e9b\u7c7b\u4f3c\u7684\u4ea7\u54c1\u6216\u76f8\u5173\u7684\u7814\u7a76\u5de5\u4f5c\uff0c\u4f46\u90fd\u6709\u4e00\u4e9b\u9650\u5236\uff1a\u6709\u7684\u7814\u7a76\u5de5\u4f5c\u53ef\u4ee5\u6839\u636e\u6587\u672c\u5408\u6210\u5339\u914d\u4e0a\u4e0b\u6587\u7684\u8bed\u97f3\uff0c\u4f46\u662f\u5fc5\u987b\u662f\u6a21\u578b\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5b66\u4e60\u8fc7\u7684\u97f3\u8272\uff1b\u6709\u7684\u4ea7\u54c1\u60f3\u5408\u6210\u5b9a\u5236\u5316\u7684\u58f0\u97f3\uff0c\u6bd4\u5982\u7528\u6237\u81ea\u5df1\u7684\u97f3\u8272\uff0c\u4f46\u9700\u8981\u7528\u6237\u51c6\u5907\u81f3\u5c1110\u5206\u949f\u7684\u58f0\u97f3\uff0c\u5e76\u5c06\u58f0\u97f3\u4e0a\u4f20\uff0c\u7136\u540e\u518d\u7b49\u5f852-24\u5c0f\u65f6\uff0c\u901a\u8fc7\u540e\u53f0\u5bf9\u58f0\u97f3\u8fdb\u884c\u8bad\u7ec3\u4e4b\u540e\uff0c\u8f6f\u4ef6\u624d\u53ef\u4ee5\u5408\u6210\u5b9a\u5236\u5316\u7684\u58f0\u97f3\u3002\u8fd9\u4e9b\u9650\u5236\u65e0\u7591\u90fd\u7ed9\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u5728\u73b0\u5b9e\u4e2d\u7684\u4f7f\u7528\u5e26\u6765\u4e86\u6781\u5927\u7684\u4e0d\u4fbf\u3002\u4e3a\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u7814\u53d1\u4e86\u4e00\u4e2a\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\uff0c\u6765\u89e3\u51b3\u8fd9\u4e9b\u6280\u672f\u96be\u70b9\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u6280\u672f\u96be\u70b9\">\u6280\u672f\u96be\u70b9<\/h2>\n\n\n\n<p>\u5728\u4ee5\u8bed\u97f3\u4e3a\u4e3b\u7684\u97f3\u89c6\u9891\u4e2d\uff0c\u8bed\u97f3\u4e2d\u7684\u5185\u5bb9\u548c\u6587\u672c\u6709\u7740\u65f6\u95f4\u4e0a\u7684\u4e00\u4e00\u5bf9\u5e94\u3002\u7814\u7a76\u5458\u4eec\u53d1\u73b0\uff0c\u82e5\u8981\u8ba9\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u53ef\u4ee5\u76f4\u63a5\u7f16\u8f91\u6587\u672c\uff0c\u518d\u6839\u636e\u8bed\u97f3\u548c\u6587\u672c\u7684\u5bf9\u5e94\u5173\u7cfb\u81ea\u52a8\u5b8c\u6210\u8bed\u97f3\u7684\u7f16\u8f91\uff0c\u9700\u8981\u7740\u91cd\u5173\u6ce8\u4ee5\u4e0b\u6280\u672f\u8981\u70b9\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"309\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-1-1024x309.png\" alt=\"\u56fe1\uff1a\u8bed\u97f3\u548c\u6587\u672c\u7684\u5bf9\u5e94\u5173\u7cfb\" class=\"wp-image-1070598\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-1-1024x309.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-1-300x91.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-1-768x232.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-1-240x72.png 240w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-1.png 1514w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe1\uff1a\u8bed\u97f3\u548c\u6587\u672c\u7684\u5bf9\u5e94\u5173\u7cfb<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>1. \u81ea\u52a8\u8bed\u97f3\u8bc6\u522b\uff1a\u5982\u679c\u8bed\u97f3\u4e0d\u662f\u6309\u7167\u5df2\u6709\u811a\u672c\u8bfb\u7684\uff0c\u90a3\u5c31\u6ca1\u6709\u6587\u672c\u4fe1\u606f\uff0c\u9700\u8981 ASR\uff08\u81ea\u52a8\u8bed\u97f3\u8bc6\u522b\uff09\u6765\u8bc6\u522b\u5f97\u5230\u6587\u672c\uff0c\u73b0\u6709\u7684 ASR \u7cfb\u7edf\u5df2\u7ecf\u80fd\u591f\u51c6\u786e\u5730\u8bc6\u522b\u8bed\u97f3\uff0c\u4f46\u53d7\u9650\u4e8e\u8bad\u7ec3\u6570\u636e\uff0c\u90e8\u5206 ASR \u7cfb\u7edf\u5e76\u4e0d\u80fd\u5b8c\u5168\u68c0\u6d4b\u51fa\u8bed\u97f3\u4e2d\u7684\u586b\u5145\u8bcd\u3002<\/p>\n\n\n\n<p>2. \u8bed\u97f3\u548c\u6587\u672c\u5bf9\u9f50\uff1a\u7814\u7a76\u4e2d\u9700\u8981\u4e00\u4e2a\u8bed\u97f3\u6587\u672c\u5bf9\u9f50\uff08forced alignment\uff09\u6a21\u5757\u6765\u63d0\u4f9b\u51c6\u786e\u7684\u6587\u672c\u548c\u8bed\u97f3\u7684\u5bf9\u9f50\u7ed3\u679c\uff0c\u4ee5\u4fbf\u53ef\u4ee5\u7cbe\u786e\u5b9a\u4f4d\u5230\u8981\u7f16\u8f91\u7684\u6587\u672c\u5728\u8bed\u97f3\u4e2d\u7684\u65f6\u95f4\u6233\u3002\u8fd9\u662f\u4e00\u4e2a\u975e\u5e38\u57fa\u7840\u4f46\u5341\u5206\u91cd\u8981\u7684\u6280\u672f\u70b9\uff0c\u4e5f\u662f\u4e00\u4e2a\u4f20\u7edf\u7684\u7814\u7a76\u95ee\u9898\uff0c\u4e00\u822c\u7528\u4e8e\u8bed\u8a00\u5b66\u7814\u7a76\u6216\u53d1\u97f3\u8bc4\u4f30\u6216\u4e3a TTS\uff08\u8bed\u97f3\u5408\u6210\uff09\u63d0\u4f9b\u5bf9\u9f50\u7684\u8bad\u7ec3\u6570\u636e\u3002\u73b0\u6709\u7684\u5f3a\u5236\u5bf9\u9f50\u65b9\u6cd5\u51e0\u4e4e\u53ef\u4ee5\u6ee1\u8db3\u8fd9\u4e9b\u9700\u6c42\uff0c\u4f46\u662f\u5bf9\u4e8e\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\uff0c\u5219\u9700\u8981\u66f4\u51c6\u786e\u7684\u5bf9\u9f50\u65b9\u5f0f\u3002\u73b0\u6709\u7684\u5bf9\u9f50\u65b9\u6cd5\u5728\u51e0\u5341\u6beb\u79d2\u7684\u8bef\u5dee\u7ea7\u522b\u4e0b\u4ecd\u4e0d\u80fd\u505a\u5230\u5b8c\u7f8e\uff0c\u800c\u4e00\u65e6\u51fa\u73b0\u51e0\u5341\u6beb\u79d2\u7684\u8bef\u5dee\uff0c\u6bd4\u5982\u5207\u5272\u8bed\u97f3\u7684\u65f6\u5019\u591a\u5207\u6216\u5c11\u5207\u4e86\u51e0\u5341\u6beb\u79d2\uff0c\u4eba\u7684\u542c\u89c9\u4f1a\u5f88\u5bb9\u6613\u5bdf\u89c9\u5230\uff0c\u5e76\u4ea7\u751f\u4e0d\u9002\u3002<\/p>\n\n\n\n<p>3. \u8bed\u97f3\u5408\u6210\uff1a\u5f53\u63d2\u5165\u6216\u4fee\u6539\u6587\u672c\u65f6\uff0c\u9700\u8981\u8bed\u97f3\u5408\u6210\u6a21\u5757\u6765\u751f\u6210\u65b0\u7684\u58f0\u97f3\u3002\u8bed\u97f3\u5408\u6210\u7684\u6700\u5927\u6311\u6218\u662f\u81ea\u7136\u548c\u6d41\u7545\uff0c\u5bf9\u4e8e\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u5c24\u5176\u91cd\u8981\uff0c\u56e0\u4e3a\u5982\u679c\u53ea\u4fee\u6539\u8bed\u97f3\u7684\u4e00\u90e8\u5206\uff0c\u7a0d\u6709\u4e0d\u8fde\u8d2f\u5c31\u4f1a\u975e\u5e38\u660e\u663e\u3002\u800c\u4e14\u7814\u7a76\u5458\u4eec\u671f\u671b\u5728\u4f7f\u7528\u8bed\u97f3\u7f16\u8f91\u6280\u672f\u7684\u65f6\u5019\uff0c\u53ef\u4ee5\u968f\u610f\u8fdb\u884c\u7f16\u8f91\u800c\u4e0d\u9700\u8981\u51c6\u5907\u8db3\u91cf\u7684\u8bed\u97f3\u6570\u636e\u53bb\u5fae\u8c03\u6a21\u578b\u3002\u56e0\u6b64\u4e00\u4e2a\u96f6\u6837\u672c\u4e0a\u4e0b\u6587\u611f\u77e5\uff08zero-shot context-aware\uff09\u7684 TTS \u662f\u5fc5\u4e0d\u53ef\u5c11\u7684\u3002<\/p>\n\n\n\n<p>4. \u586b\u5145\u8bcd\u68c0\u6d4b\uff1a\u586b\u5145\u8bcd\u68c0\u6d4b\u6a21\u5757\u53ef\u4ee5\u81ea\u52a8\u68c0\u6d4b\u8bed\u97f3\u4e2d\u7684\u586b\u5145\u8bcd\uff0c\u7528\u6237\u53ef\u4ee5\u9009\u62e9\u624b\u52a8\u5220\u9664\u90e8\u5206\u6216\u81ea\u52a8\u5168\u90e8\u5220\u9664\u3002\u4e0a\u6587\u63d0\u5230\uff0c\u90e8\u5206 ASR \u7cfb\u7edf\u5e76\u4e0d\u80fd\u68c0\u6d4b\u5230\u5168\u90e8\u586b\u5145\u8bcd\uff0c\u6240\u4ee5\u6b64\u65f6\u5c31\u6ca1\u529e\u6cd5\u901a\u8fc7\u6587\u672c\u5220\u9664\u586b\u5145\u8bcd\u6765\u7f16\u8f91\u8bed\u97f3\u3002\u6709\u7684\u8bcd\u662f\u5426\u662f\u586b\u5145\u8bcd\u53ef\u80fd\u53d6\u51b3\u4e8e\u8bed\u5883\uff0c\u6bd4\u5982\u201cyou know\u201d\u662f\u82f1\u8bed\u4e2d\u5e38\u7528\u7684\u586b\u5145\u8bcd\uff0c\u4f46\u662f\u5728\u201cDo you know him?\u201d\u8fd9\u53e5\u8bdd\u4e2d\u5b83\u5e76\u4e0d\u662f\u586b\u5145\u8bcd\uff0c\u8fd9\u65f6\u5c31\u9700\u8981\u4e00\u4e2a\u8bed\u8a00\u6a21\u578b\u6765\u8fdb\u884c\u5224\u65ad\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\">\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf<\/h2>\n\n\n\n<p>\u7814\u7a76\u5458\u4eec\u9996\u5148\u8c03\u7528\u5fae\u8f6f\u4e91\u8ba1\u7b97\u5e73\u53f0 Azure \u4e0a\u7684 ASR \u670d\u52a1\u5c06\u4e0a\u4f20\u7684\u8bed\u97f3\u6587\u4ef6\u8f6c\u5316\u4e3a\u6587\u672c\uff0c\u540c\u65f6\u8c03\u7528\u81ea\u884c\u7814\u53d1\u7684\u586b\u5145\u8bcd\u68c0\u6d4b\u6a21\u578b\uff0c\u5e76\u5c06\u586b\u5145\u8bcd\u68c0\u6d4b\u7ed3\u679c\u548c ASR \u8bc6\u522b\u7ed3\u679c\u5408\u5e76\u3002\u7136\u540e\u5c31\u53ef\u4ee5\u5bf9\u6587\u672c\u8fdb\u884c\u7f16\u8f91\u2014\u2014\u5bf9\u4e8e\u5220\u9664\u64cd\u4f5c\uff0c\u7cfb\u7edf\u4f1a\u6839\u636e\u5bf9\u9f50\u7ed3\u679c\u5220\u9664\u5bf9\u5e94\u7684\u8bed\u97f3\u7247\u6bb5\uff1b\u5bf9\u4e8e\u63d2\u5165\u64cd\u4f5c\uff0c\u7cfb\u7edf\u4f1a\u8c03\u7528\u8bed\u97f3\u5408\u6210\u6a21\u578b\u5408\u6210\u8981\u63d2\u5165\u7684\u8bed\u97f3\u5e76\u63d2\u5165\u539f\u6709\u8bed\u97f3\u4e2d\u3002\u4e0b\u9762\u662f\u51e0\u4e2a\u901a\u8fc7\u4e0a\u8ff0\u65b9\u6cd5\u5b8c\u6210\u7684\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u6837\u4f8b\uff1a<\/p>\n\n\n\n<p><strong>\u4fee\u6539\u6587\u672c\u6837\u4f8b<\/strong><\/p>\n\n\n\n<p>\u539f\u59cb\u6587\u672c\uff1aunderstand for the question and answer benchmark we&#8217;re also the first reach human parity<\/p>\n\n\n\n<p>\u4fee\u6539\u540e\u6587\u672c\uff1aunderstand for the question and answer benchmark we&#8217;re also the second reach human parity<\/p>\n\n\n\n<p>\u539f\u59cb\u8bed\u97f3\uff1a <\/p>\n\n\n\n<figure class=\"wp-block-audio\"><audio controls src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/\u4fee\u6539\u6587\u672c-\u539f\u59cb\u8bed\u97f3.mp3\"><\/audio><\/figure>\n\n\n\n<p>\u4fee\u6539\u540e\u8bed\u97f3\uff1a <\/p>\n\n\n\n<figure class=\"wp-block-audio\"><audio controls src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/\u4fee\u6539\u6587\u672c-\u4fee\u6539\u540e\u8bed\u97f3.mp3\"><\/audio><\/figure>\n\n\n\n<p><strong>\u63d2\u5165\u6587\u672c\u6837\u4f8b<\/strong><\/p>\n\n\n\n<p>\u539f\u59cb\u6587\u672c\uff1aThe song of the wretched<\/p>\n\n\n\n<p>\u4fee\u6539\u540e\u6587\u672c\uff1aThe famous song of the wretched<\/p>\n\n\n\n<p>\u539f\u59cb\u8bed\u97f3\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-audio\"><audio controls src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/\u63d2\u5165\u6587\u672c-\u539f\u59cb\u8bed\u97f3.mp3\"><\/audio><\/figure>\n\n\n\n<p>\u4fee\u6539\u540e\u8bed\u97f3\uff1a <\/p>\n\n\n\n<figure class=\"wp-block-audio\"><audio controls src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/\u63d2\u5165\u6587\u672c-\u4fee\u6539\u540e\u8bed\u97f3.mp3\"><\/audio><\/figure>\n\n\n\n<p><strong>\u5220\u9664\u6587\u672c\u6837\u4f8b<\/strong><\/p>\n\n\n\n<p>\u539f\u59cb\u6587\u672c\uff1asome have accepted it as a miracle without physical explanation<\/p>\n\n\n\n<p>\u4fee\u6539\u540e\u6587\u672c\uff1asome have accepted it without physical explanation<\/p>\n\n\n\n<p>\u539f\u59cb\u8bed\u97f3\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-audio\"><audio controls src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/\u5220\u9664\u6587\u672c-\u539f\u59cb\u8bed\u97f3.mp3\"><\/audio><\/figure>\n\n\n\n<p>\u4fee\u6539\u540e\u8bed\u97f3\uff1a <\/p>\n\n\n\n<figure class=\"wp-block-audio\"><audio controls src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/\u5220\u9664\u6587\u672c-\u4fee\u6539\u540e\u8bed\u97f3.mp3\"><\/audio><\/figure>\n\n\n\n<p><strong>\u586b\u5145\u8bcd\u68c0\u6d4b\u548c\u53bb\u9664\u6837\u4f8b<\/strong><\/p>\n\n\n\n<p>\u539f\u59cb\u6587\u672c\uff1aWe can edit your speech uh by just editing. You know, its transcript.<\/p>\n\n\n\n<p>\u4fee\u6539\u540e\u6587\u672c\uff1aWe can edit your speech by just editing its transcript.<\/p>\n\n\n\n<p>\u539f\u59cb\u8bed\u97f3\uff1a <\/p>\n\n\n\n<figure class=\"wp-block-audio\"><audio controls src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/\u586b\u5145\u8bcd\u68c0\u6d4b\u548c\u53bb\u9664-\u539f\u59cb\u8bed\u97f3.mp3\"><\/audio><\/figure>\n\n\n\n<p>\u4fee\u6539\u540e\u8bed\u97f3\uff1a <\/p>\n\n\n\n<figure class=\"wp-block-audio\"><audio controls src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/\u586b\u5145\u8bcd\u68c0\u6d4b\u548c\u53bb\u9664-\u4fee\u6539\u540e\u8bed\u97f3.mp3\"><\/audio><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u5173\u952e\u6280\u672f\u70b9\u4e00-\u8bed\u97f3\u5408\u6210\">\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u5173\u952e\u6280\u672f\u70b9\u4e00\uff1a\u8bed\u97f3\u5408\u6210<\/h2>\n\n\n\n<p>\u670d\u52a1\u4e8e\u8bed\u97f3\u7f16\u8f91\u7684\u8bed\u97f3\u5408\u6210\u6a21\u578b\u9700\u8981\u505a\u5230\u4e09\u70b9\uff1a\u96f6\u6837\u672c\u3001\u81ea\u7136\u548c\u6d41\u7545\u3002\u5176\u4e2d\u81ea\u7136\u53c8\u53ef\u7ec6\u5316\u4e3a\u4e24\u70b9\u5b50\u8981\u6c42\uff1a\u751f\u6210\u4e0e\u76ee\u6807\u8bf4\u8bdd\u4eba\u76f8\u4f3c\u7684\u97f3\u8272\uff0c\u4ee5\u53ca\u8db3\u591f\u9ad8\u7684\u97f3\u8d28\u3002\u7ecf\u8fc7\u4e0d\u65ad\u63a2\u7d22\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u8fbe\u6210\u4e86\u4ee5\u4e0a\u76ee\u6807\uff0c\u5f00\u53d1\u4e86\u4e00\u4e2a\u96f6\u6837\u672c\u4e0a\u4e0b\u6587\u611f\u77e5 TTS \u6a21\u578b RetrieverTTS\uff0c\u5e76\u5728\u8bed\u97f3\u9886\u57df\u7684\u9876\u7ea7\u5b66\u672f\u4f1a\u8bae InterSpeech 2022\u4e0a\u53d1\u8868\u4e86\u8bba\u6587\u201cRetrieverTTS: Modeling Decomposed Factors for Text-Based Speech Insertion\u201d\uff08\u6b32\u4e86\u89e3\u8bba\u6587\u8be6\u60c5\uff0c\u8bf7\u67e5\u770b\uff1ahttps:\/\/arxiv.org\/pdf\/2206.13865.pdf \uff09\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u8bbe\u8ba1\u601d\u8def\">\u8bbe\u8ba1\u601d\u8def<\/h3>\n\n\n\n<p>\u4e0d\u540c\u4e8e\u5df2\u6709\u65b9\u6cd5\u4e2d\u5c06\u8bed\u97f3\u63d2\u5165\u4efb\u52a1\u89c6\u4e3a\u6587\u672c-\u8bed\u97f3\u6a21\u6001\u878d\u5408\u7684\u601d\u8def [3,5]\uff0c\u5982\u56fe2\u6240\u793a\uff0c\u7814\u7a76\u5458\u4eec\u5c06\u8bed\u97f3\u5148\u89e3\u8026\u6210\u6587\u672c\u3001\u97f5\u5f8b\uff08\u97f3\u7d20\u5e8f\u5217\u7684\u97f3\u9ad8\u3001\u97f3\u91cf\u3001\u65f6\u957f\uff09\u3001\u97f3\u8272\u3001\u98ce\u683c\u56db\u4e2a\u8981\u7d20\uff0c\u518d\u5728\u6bcf\u4e2a\u8981\u7d20\u4e0a\u8fdb\u884c\u53ef\u63a7\u7684\u7f16\u8f91\u64cd\u4f5c\uff0c\u6700\u540e\u5c06\u56db\u90e8\u5206\u5408\u6210\u4e3a\u63d2\u5165\u540e\u7684\u8bed\u97f3\u3002\u4e00\u8a00\u4ee5\u853d\u4e4b\uff0c\u5373\u201c\u5148\u89e3\u8026\u518d\u7f16\u8f91\u201d\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"291\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-2-1024x291.png\" alt=\"\u56fe2\uff1a\u8bed\u97f3\u5408\u6210\u8bbe\u8ba1\u601d\u8def\" class=\"wp-image-1070601\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-2-1024x291.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-2-300x85.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-2-768x218.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-2-240x68.png 240w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-2.png 1431w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe2\uff1a\u8bed\u97f3\u5408\u6210\u8bbe\u8ba1\u601d\u8def<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u4f46\u662f\uff0c\u5728\u6267\u884c\u63d2\u5165\u64cd\u4f5c\u65f6\u5bf9\u56db\u79cd\u8981\u7d20\u7684\u64cd\u4f5c\u662f\u4e0d\u540c\u7684\uff1a\u6587\u672c\u53ef\u4ee5\u76f4\u63a5\u4f7f\u7528\u7528\u6237\u7f16\u8f91\u540e\u7684\u6587\u672c\uff1b\u5bf9\u4e8e\u97f5\u5f8b\u800c\u8a00\uff0c\u672a\u88ab\u7f16\u8f91\u7684\u90e8\u5206\u65e0\u9700\u6539\u53d8\uff0c\u800c\u63d2\u5165\u8bcd\u7684\u90e8\u5206\u9700\u8981\u7531\u6a21\u578b\u6839\u636e\u4e0a\u4e0b\u6587\u9884\u6d4b\u5f97\u5230\uff1b\u7531\u4e8e\u8bf4\u8bdd\u4eba\u6ca1\u6709\u6539\u53d8\uff0c\u56e0\u6b64\u97f3\u8272\u548c\u98ce\u683c\u4e24\u4e2a\u8981\u7d20\u4fdd\u6301\u4e0d\u53d8\u3002\u6587\u672c\u548c\u97f5\u5f8b\u5728\u4e00\u53e5\u8bdd\u7684\u4e0d\u540c\u65f6\u523b\u662f\u4e0d\u540c\u7684\uff0c\u800c\u97f3\u8272\u548c\u98ce\u683c\u5728\u4e00\u53e5\u8bdd\u751a\u81f3\u8fde\u7eed\u7684\u51e0\u53e5\u8bdd\u4e2d\u90fd\u4e0d\u4f1a\u6539\u53d8\uff0c\u6240\u4ee5\u524d\u8005\u4e3a\u5c40\u57df\u8981\u7d20\uff0c\u800c\u540e\u8005\u4e3a\u5168\u5c40\u8981\u7d20\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u6a21\u578b\u67b6\u6784\">\u6a21\u578b\u67b6\u6784<\/h3>\n\n\n\n<p>\u7814\u7a76\u5458\u4eec\u4f7f\u7528 Fastpitch[1] \u4f5c\u4e3a\u8bed\u97f3\u5408\u6210\u7684\u4e3b\u5e72\u7f51\u7edc\u3002\u4e3a\u4e86\u51c6\u786e\u5730\u4ee5\u96f6\u6837\u672c\u7684\u65b9\u5f0f\u9002\u5e94\u5230\u4efb\u610f\u8bf4\u8bdd\u8005\u7684\u97f3\u8272\uff0c\u5168\u5c40\u8981\u7d20\u4e0e\u5c40\u57df\u8981\u7d20\u4e4b\u95f4\u7684\u89e3\u8026\u5e94\u8db3\u591f\u7cbe\u51c6\uff0c\u5168\u5c40\u8981\u7d20\u7684\u8868\u5f81\u9700\u8db3\u591f\u5b8c\u5907\u4e14\u5e94\u6cdb\u5316\u81f3\u4efb\u610f\u8bf4\u8bdd\u4eba\u3002\u7814\u7a76\u5458\u4eec\u5728 ICLR 2022 \u53d1\u8868\u7684\u8bba\u6587\u201cRetriever: Learning Content-Style Representation as a Token-Level Bipartite Graph\u201d[2] \u4e2d\uff0c\u5df2\u7ecf\u5728\u5f88\u5927\u7a0b\u5ea6\u4e0a\u89e3\u51b3\u4e86\u8fd9\u4e00\u95ee\u9898\uff0c\u5e76\u5728\u96f6\u6837\u672c\u8bed\u97f3\u98ce\u683c\u8f6c\u6362\u4efb\u52a1\u4e2d\u53d6\u5f97\u4e86\u6700\u5148\u8fdb\u7684\u6027\u80fd\u3002\u5728\u6b64\uff0c\u7814\u7a76\u5458\u4eec\u5c06\u201cRetriever\u201d\u4e2d\u7684\u5168\u5c40\u8981\u7d20\u5efa\u6a21\u65b9\u6cd5\u5f15\u5165\u5230\u4e86\u8bed\u97f3\u63d2\u5165\u4efb\u52a1\u4e2d\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1017\" height=\"594\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-3.png\" alt=\"\u56fe3\uff1a\u8bed\u97f3\u5408\u6210\u6a21\u578b\u67b6\u6784\" class=\"wp-image-1070604\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-3.png 1017w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-3-300x175.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-3-768x449.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-3-480x280.png 480w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-3-240x140.png 240w\" sizes=\"auto, (max-width: 1017px) 100vw, 1017px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe3\uff1a\u8bed\u97f3\u5408\u6210\u6a21\u578b\u67b6\u6784<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u5b9e\u9a8c\u7ed3\u679c\">\u5b9e\u9a8c\u7ed3\u679c<\/h3>\n\n\n\n<p>\u5982\u88681\u6240\u793a\uff0cRetrieverTTS \u7684\u8bed\u97f3\u63d2\u5165\u6548\u679c\u5bf9\u63d2\u5165\u957f\u5ea6\u5e76\u4e0d\u654f\u611f\u3002\u5bf9\u4e8e\u8bad\u7ec3\u4e2d\u6ca1\u89c1\u8fc7\u7684\u8bf4\u8bdd\u4eba\uff0c\u5373\u4f7f\u63d2\u5165\u8bed\u97f3\u957f\u5ea6\u8d85\u8fc7\u4e24\u79d2\u4e43\u81f3\u751f\u6210\u4e00\u6574\u53e5\uff08long insert, full generation\uff09\uff0c\u5176\u8bed\u97f3\u81ea\u7136\u5ea6\u4ecd\u7136\u4fdd\u6301\u5728\u8f83\u9ad8\u6c34\u5e73\u3002\u5728\u63d2\u5165\u5c11\u4e8e6\u4e2a\u8bcd\u65f6 \uff08short insert, mid insert\uff09\uff0c\u751a\u81f3\u8fbe\u5230\u4e0e\u771f\u4eba\u5f55\u97f3\u76f8\u8fd1\u7684\u81ea\u7136\u5ea6\u8bc4\u5206\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"654\" height=\"260\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-4.png\" alt=\"\u88681\uff1a\u8bed\u97f3\u5408\u6210\u5bf9\u4e0d\u540c\u63d2\u5165\u957f\u5ea6\u7684\u9c81\u68d2\u6027\u6d4b\u8bd5\" class=\"wp-image-1070607\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-4.png 654w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-4-300x119.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-4-240x95.png 240w\" sizes=\"auto, (max-width: 654px) 100vw, 654px\" \/><figcaption class=\"wp-element-caption\"><em>\u88681\uff1a\u8bed\u97f3\u5408\u6210\u5bf9\u4e0d\u540c\u63d2\u5165\u957f\u5ea6\u7684\u9c81\u68d2\u6027\u6d4b\u8bd5<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u5728\u88682\u7684\u6d88\u878d\u5b9e\u9a8c\u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u5206\u522b\u53bb\u6389\u4e86\u5bf9\u6297\u8bad\u7ec3\uff08- adv\uff09\uff0c\u97f5\u5f8b\u5e73\u6ed1\u4efb\u52a1\uff08- prosody-smooth\uff09\uff0c\u4ee5\u53ca Retriever \u7684\u5168\u5c40\u8981\u7d20\u5efa\u6a21\u65b9\u6cd5 \uff08- retriever\uff09\uff0c\u7ed3\u679c\u5747\u53d1\u73b0\u660e\u663e\u7684\u6027\u80fd\u4e0b\u964d\u3002\u5728\u4e09\u6b21\u5b9e\u9a8c\u7684\u6d4b\u8bd5\u6837\u4f8b\u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u5206\u522b\u53d1\u73b0\u4e86\u97f3\u8d28\u5dee\u3001\u97f5\u5f8b\u4e0d\u8fde\u8d2f\u4ee5\u53ca\u97f3\u8272\u4e0d\u50cf\u7684\u95ee\u9898\u3002\u8fd9\u9a8c\u8bc1\u4e86 RetrieverTTS \u4e09\u65b9\u9762\u8bbe\u8ba1\u5747\u8fbe\u5230\u4e86\u8bbe\u8ba1\u521d\u8877\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"775\" height=\"312\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-5.png\" alt=\"\u88682\uff1a\u8bed\u97f3\u5408\u6210\u6d88\u878d\u5b9e\u9a8c\" class=\"wp-image-1070610\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-5.png 775w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-5-300x121.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-5-768x309.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-5-240x97.png 240w\" sizes=\"auto, (max-width: 775px) 100vw, 775px\" \/><figcaption class=\"wp-element-caption\"><em>\u88682\uff1a\u8bed\u97f3\u5408\u6210\u6d88\u878d\u5b9e\u9a8c<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u5728\u88683\u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u4e0e\u5176\u4ed6\u65b9\u6cd5\u8fdb\u884c\u4e86\u5bf9\u6bd4\uff0c\u53d1\u73b0\u57fa\u4e8e\u6a21\u6001\u878d\u5408\u7684\u65b9\u6cd5[3]\u5728\u63d2\u5165\u8f83\u957f\u8bed\u97f3\u7684\u60c5\u5f62\u4e0b\u8bed\u97f3\u81ea\u7136\u5ea6\u8868\u73b0\u8f83\u5dee\uff0c\u800c\u5176\u4ed6\u7684\u96f6\u6837\u672c\u8bf4\u8bdd\u4eba\u81ea\u9002\u5e94\u8bed\u97f3\u5408\u6210\uff08zero-shot speaker adaptive TTS\uff09[4] \u5728\u97f3\u8272\u76f8\u4f3c\u5ea6\u65b9\u9762\u4e0e RetrieverTTS \u7684\u65b9\u6cd5\u6709\u8f83\u5927\u5dee\u8ddd\u3002\u4e0a\u8ff0\u5bf9\u6bd4\u4f53\u73b0\u51fa\u4e86 RetrieverTTS \u65b9\u6cd5\u7684\u4f18\u8d8a\u6027\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"690\" height=\"224\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-6.png\" alt=\"\u88683\uff1a\u8bed\u97f3\u5408\u6210\u7cfb\u7edf\u5bf9\u6bd4\" class=\"wp-image-1070613\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-6.png 690w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-6-300x97.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-6-240x78.png 240w\" sizes=\"auto, (max-width: 690px) 100vw, 690px\" \/><figcaption class=\"wp-element-caption\"><em>\u88683\uff1a\u8bed\u97f3\u5408\u6210\u7cfb\u7edf\u5bf9\u6bd4<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u5173\u952e\u6280\u672f\u70b9\u4e8c-\u586b\u5145\u8bcd\u68c0\u6d4b\">\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u5173\u952e\u6280\u672f\u70b9\u4e8c\uff1a\u586b\u5145\u8bcd\u68c0\u6d4b<\/h2>\n\n\n\n<p>\u5f88\u591a ASR \u6a21\u578b\u7531\u4e8e\u53d7\u9650\u4e8e\u8bad\u7ec3\u6570\u636e\uff0c\u4e0d\u80fd\u5b8c\u6574\u7684\u68c0\u6d4b\u5230\u586b\u5145\u8bcd\uff0c\u56e0\u6b64\u9700\u8981\u4e00\u4e2a\u5355\u72ec\u7684\u6a21\u5757\u8fdb\u884c\u586b\u5145\u8bcd\u7684\u68c0\u6d4b\u3002\u4e8b\u5b9e\u4e0a\uff0c\u57fa\u4e8e\u8bed\u97f3\u7684\u586b\u5145\u8bcd\u68c0\u6d4b\u6280\u672f\u5c5e\u4e8e\u8bed\u97f3\u5173\u952e\u8bcd\u68c0\u6d4b\u7684\u4e00\u4e2a\u7279\u4f8b\uff0c\u6240\u4ee5\u7814\u7a76\u5458\u4eec\u5c06\u8bed\u97f3\u5173\u952e\u8bcd\u68c0\u6d4b\u89c6\u4e3a\u76ee\u6807\u68c0\u6d4b\u95ee\u9898\uff0c\u800c\u4e0d\u662f\u8bed\u97f3\u5206\u7c7b\u95ee\u9898\u3002\u53d7\u8ba1\u7b97\u673a\u89c6\u89c9\u4e2d\u76ee\u6807\u68c0\u6d4b\u65b9\u6cd5\u7684\u542f\u53d1[6], \u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u540d\u4e3a AF-KWS\uff08anchor free detector for continuous speech keyword spotting\uff09 \u7684\u5173\u952e\u8bcd\u68c0\u6d4b\u65b9\u6cd5\u3002<\/p>\n\n\n\n<p>\u5728 AF-KWS \u65b9\u6cd5\u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u901a\u8fc7\u9884\u6d4b\u4e00\u4e2a\u5173\u952e\u8bcd\u70ed\u529b\u56fe\uff0c\u5f97\u5230\u6bcf\u4e00\u7c7b\u5173\u952e\u8bcd\u5728\u8fde\u7eed\u8bed\u97f3\u4e2d\u7684\u4f4d\u7f6e\uff0c\u7136\u540e\u901a\u8fc7\u4e24\u4e2a\u9884\u6d4b\u6a21\u5757\uff0c\u5206\u522b\u9884\u6d4b\u5173\u952e\u8bcd\u7684\u957f\u5ea6\u548c\u7528\u4e8e\u77eb\u6b63\u5173\u952e\u8bcd\u4f4d\u7f6e\u8bef\u5dee\u7684\u4f4d\u7f6e\u504f\u79fb\u91cf\u3002\u4e0d\u540c\u4e8e\u8ba1\u7b97\u673a\u89c6\u89c9\u4e2d\u7684\u76ee\u6807\u68c0\u6d4b\u7b97\u6cd5[6][7][8]\uff0c\u7814\u7a76\u5458\u4eec\u5f15\u5165\u4e86\u4e00\u4e2a\u201cunknown\u201d\u7c7b\u522b\uff0c\u8868\u793a\u975e\u76ee\u6807\u5173\u952e\u8bcd\u7684\u5176\u4ed6\u8bcd\uff0c\u8fd9\u79cd\u8bbe\u8ba1\u5c06\u201cunknown\u201d\u548c\u8bed\u97f3\u4e2d\u7684\u80cc\u666f\u566a\u97f3\u548c\u5b89\u9759\u7247\u6bb5\u5206\u5f00\uff0c\u80fd\u591f\u663e\u8457\u63d0\u9ad8\u5173\u952e\u8bcd\u68c0\u6d4b\u7684\u51c6\u786e\u6027\u3002\u8be5\u65b9\u6cd5\u7684\u8bba\u6587\u201cAn Anchor-Free Detector for Continuous Speech Keyword Spotting\u201d\u5df2\u7ecf\u88ab InterSpeech 2022 \u63a5\u6536\uff08\u66f4\u591a\u8bba\u6587\u7ec6\u8282\uff0c\u8bf7\u67e5\u770b\uff1ahttps:\/\/arxiv.org\/pdf\/2208.04622.pdf \uff09\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u7b97\u6cd5\u6846\u67b6\">\u7b97\u6cd5\u6846\u67b6<\/h3>\n\n\n\n<p>\u5982\u56fe4\u6240\u793a\uff0c\u5bf9\u4e8e\u8f93\u5165\u8bed\u97f3\uff0c\u7814\u7a76\u5458\u4eec\u9996\u5148\u63d0\u53d6\u4e86\u8bed\u97f3\u7684 STFT \u9891\u8c31\u56fe\uff0c\u7136\u540e\u4f7f\u7528 ResNet[9] \u8fdb\u4e00\u6b65\u63d0\u53d6\u7279\u5f81\uff0c\u7136\u540e\u5c06\u7279\u5f81\u8f93\u5165\u4e09\u4e2a\u9884\u6d4b\u6a21\u5757\uff0c\u5206\u522b\u7528\u4e8e\u9884\u6d4b\u5173\u952e\u8bcd\u7684\u70ed\u529b\u56fe\u3001\u5173\u952e\u8bcd\u7684\u957f\u5ea6\u548c\u5173\u952e\u8bcd\u4f4d\u7f6e\u504f\u7f6e\u3002\u5728\u8bad\u7ec3\u9636\u6bb5\uff0c\u70ed\u529b\u56fe\u4ee5\u5173\u952e\u8bcd\u7684\u4f4d\u7f6e\u4e3a\u4e2d\u5fc3\uff0c\u4f7f\u7528\u9ad8\u65af\u6838\u51fd\u6570\u5c06\u5173\u952e\u8bcd\u7684\u4f4d\u7f6e\u6269\u5c55\u3002\u5728\u9884\u6d4b\u9636\u6bb5\uff0c\u7814\u7a76\u5458\u4eec\u53d6\u9884\u6d4b\u5f97\u5230\u7684\u70ed\u529b\u56fe\u7684\u5cf0\u503c\u70b9\u4f5c\u4e3a\u9884\u6d4b\u5f97\u5230\u7684\u5173\u952e\u8bcd\u7684\u4f4d\u7f6e\uff0c\u7136\u540e\u63d0\u53d6\u5bf9\u5e94\u4f4d\u7f6e\u7684\u5173\u952e\u8bcd\u957f\u5ea6\u548c\u504f\u7f6e\u7684\u9884\u6d4b\u7ed3\u679c\uff0c\u8ba1\u7b97\u5f97\u5230\u6700\u7ec8\u7684\u5173\u952e\u8bcd\u7684\u4f4d\u7f6e\u548c\u7c7b\u522b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"209\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-7-1024x209.png\" alt=\"\u56fe4\uff1aAF-KWS \u7b97\u6cd5\u6846\u67b6\" class=\"wp-image-1070616\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-7-1024x209.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-7-300x61.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-7-768x157.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-7-1536x314.png 1536w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-7-2048x419.png 2048w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-7-240x49.png 240w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe4\uff1aAF-KWS \u7b97\u6cd5\u6846\u67b6<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u5b9e\u9a8c\u7ed3\u8bba\">\u5b9e\u9a8c\u7ed3\u8bba<\/h3>\n\n\n\n<p>\u7814\u7a76\u5458\u4eec\u9009\u7528\u4e86\u4e24\u4e2a\u5148\u8fdb\u7684\u5173\u952e\u8bcd\u68c0\u6d4b\u6a21\u578b[10][11]\u505a\u5bf9\u6bd4\uff0c\u5728\u8fde\u7eed\u8bed\u97f3\u5173\u952e\u8bcd\u68c0\u6d4b\u7684\u5b9e\u9a8c\u8bbe\u5b9a\u4e2d\uff0cAF-KWS \u7684\u6a21\u578b\u5728\u53ef\u6bd4\u7684\u8fd0\u884c\u901f\u5ea6\u4e0b\uff0c\u5e73\u5747\u51c6\u786e\u7387\u8fdc\u8d85\u5176\u4ed6\u6a21\u578b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"143\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-8-1024x143.png\" alt=\"\u88684\uff1a\u586b\u5145\u8bcd\u68c0\u6d4b\u6027\u80fd\u5bf9\u6bd4\" class=\"wp-image-1070619\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-8-1024x143.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-8-300x42.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-8-768x107.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-8-240x34.png 240w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-8.png 1116w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u88684\uff1a\u586b\u5145\u8bcd\u68c0\u6d4b\u6027\u80fd\u5bf9\u6bd4<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u4e3a\u4e86\u9a8c\u8bc1 AF-KWS \u65b9\u6cd5\u7684\u63d0\u5347\u4e0d\u662f\u56e0\u4e3a backbone \u66f4\u5f3a\u5927\uff0c\u7814\u7a76\u5458\u4eec\u5c06\u4e09\u4e2a\u9884\u6d4b\u6a21\u5757\u66ff\u6362\u6210\u4e86\u4e00\u4e2a\u5206\u7c7b\u6a21\u5757\uff08AF-KWS-cls\uff09\uff0c\u53d1\u73b0\u6a21\u578b\u6027\u80fd\u660e\u663e\u4e0b\u964d\u3002\u4e3a\u4e86\u9a8c\u8bc1\u5f15\u5165\u7684\u201cunknown\u201d\u7c7b\u522b\u7684\u6709\u6548\u6027\uff0c\u7814\u7a76\u5458\u4eec\u53bb\u6389\u4e86\u8fd9\u4e2a\u7c7b\u522b(w\/o unknown)\uff0c\u53d1\u73b0\u6a21\u578b\u6027\u80fd\u4e5f\u660e\u663e\u4e0b\u964d\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"301\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-9-1024x301.png\" alt=\"\u88685\uff1a\u586b\u5145\u8bcd\u68c0\u6d4b\u6d88\u878d\u5b9e\u9a8c\" class=\"wp-image-1070622\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-9-1024x301.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-9-300x88.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-9-768x226.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-9-240x71.png 240w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/text-based-speech-editing-9.png 1071w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u88685\uff1a\u586b\u5145\u8bcd\u68c0\u6d4b\u6d88\u878d\u5b9e\u9a8c<\/em><\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u5173\u952e\u8bcd\u68c0\u6d4b\u6a21\u578b\u7528\u4e8e\u586b\u5145\u8bcd\u68c0\u6d4b\">\u5173\u952e\u8bcd\u68c0\u6d4b\u6a21\u578b\u7528\u4e8e\u586b\u5145\u8bcd\u68c0\u6d4b<\/h3>\n\n\n\n<p>\u7531\u4e8e\u586b\u5145\u8bcd\u4e5f\u53ef\u4ee5\u770b\u4f5c\u4e00\u79cd\u7279\u6b8a\u7684\u5173\u952e\u8bcd\uff0c\u6240\u4ee5\u7814\u7a76\u5458\u4eec\u57fa\u4e8e SwitchBoard \u6570\u636e\u96c6[12]\uff0c\u5236\u4f5c\u4e86\u4e00\u4e2a\u8bed\u97f3\u586b\u5145\u8bcd\u68c0\u6d4b\u6570\u636e\u96c6\uff0c\u5e76\u5728\u8fd9\u4e2a\u6570\u636e\u96c6\u4e0a\u91cd\u65b0\u8bad\u7ec3\u5173\u952e\u8bcd\u68c0\u6d4b\u6a21\u578b\u3002\u5728\u771f\u5b9e\u7684\u6d4b\u8bd5\u6570\u636e\u4e2d\uff0cAF-KWS \u65b9\u6cd5\u5f97\u5230\u4e86\u4e0e\u5e02\u9762\u4e0a\u6700\u597d\u7684\u65b9\u6cd5\u51e0\u4e4e\u76f8\u540c\u7684\u6027\u80fd\u3002\u9488\u5bf9\u586b\u5145\u8bcd\u7684\u7279\u70b9\uff0c\u6bd4\u5982\u586b\u5145\u8bcd\u4e00\u822c\u5305\u542b\u7684\u97f3\u8282\u8f83\u5c11\uff0c\u66f4\u5bb9\u6613\u4e0e\u7279\u5b9a\u7c7b\u522b\u7684\u8bcd\u6df7\u6dc6\uff0c\u7814\u7a76\u5458\u4eec\u4f1a\u7ee7\u7eed\u6539\u8fdb\u6a21\u578b\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u672a\u6765\u5c55\u671b\">\u672a\u6765\u5c55\u671b<\/h2>\n\n\n\n<p>\u5c3d\u7ba1\u73b0\u6709\u7684\u6280\u672f\u548c\u672c\u6587\u4e2d\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u5df2\u7ecf\u5b9e\u73b0\u4e86\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7684\u90e8\u5206\u529f\u80fd\uff0c\u4f46\u4ecd\u6709\u5f88\u591a\u7814\u7a76\u9700\u8981\u6301\u7eed\u63a2\u7d22\uff0c\u5176\u4e2d\u5305\u62ec\uff1a\u5f00\u53d1\u5bcc\u6587\u672c\u683c\u5f0f\uff0c\u8fdb\u884c\u8bed\u97f3\u89e3\u8026\uff0c\u7cbe\u51c6\u63a7\u5236\u8bed\u97f3\u7684\u91cd\u5ea6\uff0c\u8bed\u6c14\u8bed\u8c03\u548c\u60c5\u7eea\uff1b\u5f00\u53d1\u66f4\u7cbe\u786e\u7684\u8bed\u97f3\u6587\u672c\u5bf9\u9f50\u7b97\u6cd5\uff1b\u5728 TTS \u4e2d\u80cc\u666f\u566a\u97f3\u6216\u80cc\u666f\u97f3\u4e50\u8fdb\u884c\u5efa\u6a21\uff0c\u8ba9\u5408\u6210\u7684\u8bed\u97f3\u5305\u542b\u5339\u914d\u7684\u80cc\u666f\u566a\u58f0\u6216\u80cc\u666f\u97f3\u4e50\uff1b\u5f00\u53d1\u7ed3\u5408\u8bed\u97f3\u548c\u6587\u672c\u7684\u591a\u6a21\u6001\u586b\u5145\u8bcd\u68c0\u6d4b\u68c0\u6d4b\u7b97\u6cd5\uff1b\u7b49\u7b49\u3002<\/p>\n\n\n\n<p>\u53c2\u8003\u6587\u732e\uff1a<\/p>\n\n\n\n<p>[1] A. Lancucki, \u201cFastpitch: Parallel text-to-speech with pitch prediction,\u201d in ICASSP, 2021.<\/p>\n\n\n\n<p>[2] Y. Dacheng, R. Xuanchi, L. Chong, W. Yuwang, X. Zhiwei, and Z. Wenjun, \u201cRetriever: Learning content-style representation as a token-level bipartite graph,\u201d in ICLR, 2022.<\/p>\n\n\n\n<p>[3] C. Tang, C. Luo, Z. Zhao, D. Yin, Y. Zhao, and W. Zeng, \u201cZero-shot text-to-speech for text-based insertion in audio narration,\u201d in Interspeech, 2021.<\/p>\n\n\n\n<p>[4] D. Min, D. B. Lee, E. Yang, and S. J. Hwang, \u201cMeta-stylespeech : Multi-speaker adaptive text-to-speech generation,\u201d in ICML, 2021.<\/p>\n\n\n\n<p>[5] Z. Borsos, M. Sharifi, and M. Tagliasacchi, \u201cSpeechpainter: Textconditioned speech inpainting,\u201d arXiv preprint arXiv:2202.07273, 2022.<\/p>\n\n\n\n<p>[6] X. Zhou, D. Wang, and P. Kr\u00a8ahenb\u00a8uhl, \u201cObjects as points,\u201d arXiv preprint arXiv:1904.07850, 2019.<\/p>\n\n\n\n<p>[7] C. Zhu, Y. He, and M. Savvides, \u201cFeature selective anchor-free module for single-shot object detection,\u201d in Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition, 2019, pp. 840\u2013849.<\/p>\n\n\n\n<p>[8] T. Kong, F. Sun, H. Liu, Y. Jiang, L. Li, and J. Shi, \u201cFoveabox: Beyound anchor-based object detection,\u201d IEEE Transactions on Image Processing, vol. 29, pp. 7389\u20137398, 2020.<\/p>\n\n\n\n<p>[9] B. Xiao, H. Wu, and Y. Wei, \u201cSimple baselines for human pose estimation and tracking,\u201d in Proceedings of the European conference on computer vision (ECCV), 2018, pp. 466\u2013481.<\/p>\n\n\n\n<p>[10] O. Rybakov, N. Kononenko, N. Subrahmanya, M. Visontai, and S. Laurenzo, \u201cStreaming keyword spotting on mobile devices,\u201d Proc. Interspeech 2020, pp. 2277\u20132281, 2020.<\/p>\n\n\n\n<p>[11] S. Majumdar and B. Ginsburg, \u201cMatchboxnet: 1d time-channel separable convolutional neural network architecture for speech commands recognition,\u201d Proc. Interspeech 2020, pp. 3356\u20133360, 2020.<\/p>\n\n\n\n<p>[12] John J Godfrey, Edward C Holliman, and Jane Mc-Daniel, \u201cSwitchboard: Telephone speech corpus for research and development,\u201d in Acoustics, Speech, and Signal Processing, 1992. ICASSP-92., 1992 IEEE International Conference on. IEEE, 1992, vol. 1, pp. 517\u2013520.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u4f5c\u8005\uff1a\u8d75\u4e4b\u6e90\u3001\u7f57\u7fc0 \u7f16\u8005\u6309\uff1a\u5982\u4eca\u5728\u5404\u79cd\u793e\u4ea4\u7f51\u7edc\u5e73\u53f0\u4e0a\u53d1\u5e03\u7684\u89c6\u9891\uff0c\u56e0\u62cd\u6444\u4fbf\u6377\u3001\u53ef\u5b9e\u65f6\u5206\u4eab\u3001\u4e92\u52a8\u4ea4\u6d41\u7b49\u7279\u70b9\u800c\u6df1\u53d7\u5927\u4f17\u559c\u7231\u3002\u89c6\u9891\u6df1\u523b\u5f71\u54cd\u548c\u6539\u53d8\u4e86\u4eba\u4eec\u89c2\u5bdf\u4e16\u754c\u3001\u8bb0\u5f55\u751f\u6d3b\u548c\u8868\u8fbe\u60c5\u611f\u7684\u65b9\u5f0f\u3002\u7136\u800c\uff0c\u73b0\u5728\u5e02\u9762\u4e0a\u8bb8\u591a\u89c6\u9891\u6216\u97f3\u9891\u526a\u8f91\u8f6f\u4ef6\u4e3a\u4e86\u6ee1\u8db3\u7528\u6237\u9700\u6c42\u5c3d\u7ba1\u62e5\u6709\u4e30\u5bcc\u7684\u529f\u80fd\uff0c\u4f46\u64cd\u4f5c\u5374\u5f88\u590d\u6742\uff0c\u5f88\u591a\u7b80\u5355\u7684\u526a\u8f91\u4efb\u52a1\u90fd\u8fd8\u9700\u8981\u5728\u8f6f\u4ef6\u4e2d\u9010\u5e27\u5bf9\u7167\u786e\u5b9a\u526a\u5207\u65f6\u95f4\u70b9\u3002\u5bf9\u4e8e\u4ee5\u8bed\u97f3\u4e3a\u4e3b\u8981\u80cc\u666f\u58f0\u97f3\u7684\u89c6\u9891\uff0c\u5982\u7ebf\u4e0a\u4f1a\u8bae\u5f55\u50cf\u3001\u6f14\u793a\u89c6\u9891\u3001vlog \u7b49\uff0c\u5982\u679c\u6211\u4eec\u80fd\u901a\u8fc7\u7f16\u8f91\u6587\u672c\u7684\u5f62\u5f0f\uff0c\u76f4\u63a5\u7f16\u8f91\u97f3\u89c6\u9891\u4e2d\u7684\u8bed\u97f3\u5185\u5bb9\uff0c\u8ba9\u97f3\u89c6\u9891\u7684\u7f16\u8f91\u81ea\u52a8\u6839\u636e\u6587\u672c\u5b8c\u6210\uff0c\u90a3\u4e48\u5c06\u5927\u5927\u964d\u4f4e\u97f3\u89c6\u9891\u7684\u7f16\u8f91\u96be\u5ea6\uff0c\u63d0\u9ad8\u521b\u4f5c\u8005\u7684\u6548\u7387\u3002\u4e3a\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u7814\u53d1\u4e86\u4e00\u4e2a\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u3002\u672c\u6587\u5c06\u8be6\u7ec6\u4ecb\u7ecd\u8fd9\u4e2a\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u548c\u7814\u7a76\u5458\u4eec\u7814\u53d1\u7684\u8bed\u97f3\u5408\u6210\u53ca\u586b\u5145\u8bcd\u68c0\u6d4b\u6280\u672f\u3002 \u65e0\u8bba\u662f\u6f14\u793a\u89c6\u9891\u3001\u6559\u5b66\u89c6\u9891\u3001\u4f1a\u8bae\u5f55\u50cf\u8fd8\u662f\u8bb0\u5f55\u751f\u6d3b\u7247\u6bb5\u7684 vlog\uff0c\u5728\u5f88\u591a\u5b9e\u9645\u7684\u5e94\u7528\u573a\u666f\u4e2d\uff0c\u4eba\u4eec\u5e38\u5e38\u9700\u8981\u91cd\u65b0\u5f55\u5236\u8bed\u97f3\uff08\u89c6\u9891\uff09\u6216\u5bf9\u8bed\u97f3\uff08\u89c6\u9891\uff09\u8fdb\u884c\u7f16\u8f91\u3002\u56e0\u4e3a\u62cd\u6444\u7684\u7d20\u6750\u4e2d\u5f80\u5f80\u4f1a\u5b58\u5728\u5927\u91cf\u505c\u987f\u548c\u8131\u53e3\u800c\u51fa\u3001\u8bcd\u4e0d\u8fbe\u610f\u7684\u8bed\u53e5\uff0c\u6216\u8005\u662f\u5197\u4f59\u7684\u5185\u5bb9\u3002\u4f46\u7531\u4e8e\u58f0\u97f3\u7684\u7279\u6027\uff0c\u6211\u4eec\u6ca1\u6709\u529e\u6cd5\u5728\u5f55\u97f3\u5e95\u672c\u7684\u57fa\u7840\u4e0a\u53bb\u4fee\u6539\u5b57\u8bcd\uff0c\u53ea\u80fd\u4e00\u5e27\u4e00\u5e27\u5728\u526a\u8f91\u4e0a\u4e0b\u529f\u592b\uff0c\u56e0\u6b64\u58f0\u97f3\u7684\u526a\u8f91\u5de5\u4f5c\u7e41\u7410\u53c8\u5145\u6ee1\u6311\u6218\u3002\u5982\u679c\u62e5\u6709\u4e00\u4e2a\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\uff0c\u53ef\u4ee5\u901a\u8fc7\u76f4\u63a5\u7f16\u8f91\u8bed\u97f3\u5bf9\u5e94\u7684\u6587\u672c\uff0c\u5b8c\u6210\u5bf9\u8bed\u97f3\uff08\u89c6\u9891\uff09\u7684\u7f16\u8f91\uff0c\u90a3\u4e48\u666e\u901a\u7528\u6237\u4e5f\u80fd\u6210\u4e3a\u4e00\u4e2a\u6709\u521b\u610f\u7684\u526a\u8f91\u5e08\uff0c\u628a\u4e00\u6bb5\u5197\u6742\u7684\u97f3\u89c6\u9891\u53d8\u5f97\u6e05\u6670\u3001\u81ea\u7136\u53c8\u4e13\u4e1a\u3002 \u5e02\u573a\u4e0a\u73b0\u5728\u5df2\u7ecf\u6709\u4e00\u4e9b\u7c7b\u4f3c\u7684\u4ea7\u54c1\u6216\u76f8\u5173\u7684\u7814\u7a76\u5de5\u4f5c\uff0c\u4f46\u90fd\u6709\u4e00\u4e9b\u9650\u5236\uff1a\u6709\u7684\u7814\u7a76\u5de5\u4f5c\u53ef\u4ee5\u6839\u636e\u6587\u672c\u5408\u6210\u5339\u914d\u4e0a\u4e0b\u6587\u7684\u8bed\u97f3\uff0c\u4f46\u662f\u5fc5\u987b\u662f\u6a21\u578b\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5b66\u4e60\u8fc7\u7684\u97f3\u8272\uff1b\u6709\u7684\u4ea7\u54c1\u60f3\u5408\u6210\u5b9a\u5236\u5316\u7684\u58f0\u97f3\uff0c\u6bd4\u5982\u7528\u6237\u81ea\u5df1\u7684\u97f3\u8272\uff0c\u4f46\u9700\u8981\u7528\u6237\u51c6\u5907\u81f3\u5c1110\u5206\u949f\u7684\u58f0\u97f3\uff0c\u5e76\u5c06\u58f0\u97f3\u4e0a\u4f20\uff0c\u7136\u540e\u518d\u7b49\u5f852-24\u5c0f\u65f6\uff0c\u901a\u8fc7\u540e\u53f0\u5bf9\u58f0\u97f3\u8fdb\u884c\u8bad\u7ec3\u4e4b\u540e\uff0c\u8f6f\u4ef6\u624d\u53ef\u4ee5\u5408\u6210\u5b9a\u5236\u5316\u7684\u58f0\u97f3\u3002\u8fd9\u4e9b\u9650\u5236\u65e0\u7591\u90fd\u7ed9\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u5728\u73b0\u5b9e\u4e2d\u7684\u4f7f\u7528\u5e26\u6765\u4e86\u6781\u5927\u7684\u4e0d\u4fbf\u3002\u4e3a\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u7814\u53d1\u4e86\u4e00\u4e2a\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\uff0c\u6765\u89e3\u51b3\u8fd9\u4e9b\u6280\u672f\u96be\u70b9\u3002 \u5728\u4ee5\u8bed\u97f3\u4e3a\u4e3b\u7684\u97f3\u89c6\u9891\u4e2d\uff0c\u8bed\u97f3\u4e2d\u7684\u5185\u5bb9\u548c\u6587\u672c\u6709\u7740\u65f6\u95f4\u4e0a\u7684\u4e00\u4e00\u5bf9\u5e94\u3002\u7814\u7a76\u5458\u4eec\u53d1\u73b0\uff0c\u82e5\u8981\u8ba9\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u7cfb\u7edf\u53ef\u4ee5\u76f4\u63a5\u7f16\u8f91\u6587\u672c\uff0c\u518d\u6839\u636e\u8bed\u97f3\u548c\u6587\u672c\u7684\u5bf9\u5e94\u5173\u7cfb\u81ea\u52a8\u5b8c\u6210\u8bed\u97f3\u7684\u7f16\u8f91\uff0c\u9700\u8981\u7740\u91cd\u5173\u6ce8\u4ee5\u4e0b\u6280\u672f\u8981\u70b9\uff1a 1. \u81ea\u52a8\u8bed\u97f3\u8bc6\u522b\uff1a\u5982\u679c\u8bed\u97f3\u4e0d\u662f\u6309\u7167\u5df2\u6709\u811a\u672c\u8bfb\u7684\uff0c\u90a3\u5c31\u6ca1\u6709\u6587\u672c\u4fe1\u606f\uff0c\u9700\u8981 ASR\uff08\u81ea\u52a8\u8bed\u97f3\u8bc6\u522b\uff09\u6765\u8bc6\u522b\u5f97\u5230\u6587\u672c\uff0c\u73b0\u6709\u7684 ASR \u7cfb\u7edf\u5df2\u7ecf\u80fd\u591f\u51c6\u786e\u5730\u8bc6\u522b\u8bed\u97f3\uff0c\u4f46\u53d7\u9650\u4e8e\u8bad\u7ec3\u6570\u636e\uff0c\u90e8\u5206 ASR \u7cfb\u7edf\u5e76\u4e0d\u80fd\u5b8c\u5168\u68c0\u6d4b\u51fa\u8bed\u97f3\u4e2d\u7684\u586b\u5145\u8bcd\u3002 2. \u8bed\u97f3\u548c\u6587\u672c\u5bf9\u9f50\uff1a\u7814\u7a76\u4e2d\u9700\u8981\u4e00\u4e2a\u8bed\u97f3\u6587\u672c\u5bf9\u9f50\uff08forced alignment\uff09\u6a21\u5757\u6765\u63d0\u4f9b\u51c6\u786e\u7684\u6587\u672c\u548c\u8bed\u97f3\u7684\u5bf9\u9f50\u7ed3\u679c\uff0c\u4ee5\u4fbf\u53ef\u4ee5\u7cbe\u786e\u5b9a\u4f4d\u5230\u8981\u7f16\u8f91\u7684\u6587\u672c\u5728\u8bed\u97f3\u4e2d\u7684\u65f6\u95f4\u6233\u3002\u8fd9\u662f\u4e00\u4e2a\u975e\u5e38\u57fa\u7840\u4f46\u5341\u5206\u91cd\u8981\u7684\u6280\u672f\u70b9\uff0c\u4e5f\u662f\u4e00\u4e2a\u4f20\u7edf\u7684\u7814\u7a76\u95ee\u9898\uff0c\u4e00\u822c\u7528\u4e8e\u8bed\u8a00\u5b66\u7814\u7a76\u6216\u53d1\u97f3\u8bc4\u4f30\u6216\u4e3a TTS\uff08\u8bed\u97f3\u5408\u6210\uff09\u63d0\u4f9b\u5bf9\u9f50\u7684\u8bad\u7ec3\u6570\u636e\u3002\u73b0\u6709\u7684\u5f3a\u5236\u5bf9\u9f50\u65b9\u6cd5\u51e0\u4e4e\u53ef\u4ee5\u6ee1\u8db3\u8fd9\u4e9b\u9700\u6c42\uff0c\u4f46\u662f\u5bf9\u4e8e\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\uff0c\u5219\u9700\u8981\u66f4\u51c6\u786e\u7684\u5bf9\u9f50\u65b9\u5f0f\u3002\u73b0\u6709\u7684\u5bf9\u9f50\u65b9\u6cd5\u5728\u51e0\u5341\u6beb\u79d2\u7684\u8bef\u5dee\u7ea7\u522b\u4e0b\u4ecd\u4e0d\u80fd\u505a\u5230\u5b8c\u7f8e\uff0c\u800c\u4e00\u65e6\u51fa\u73b0\u51e0\u5341\u6beb\u79d2\u7684\u8bef\u5dee\uff0c\u6bd4\u5982\u5207\u5272\u8bed\u97f3\u7684\u65f6\u5019\u591a\u5207\u6216\u5c11\u5207\u4e86\u51e0\u5341\u6beb\u79d2\uff0c\u4eba\u7684\u542c\u89c9\u4f1a\u5f88\u5bb9\u6613\u5bdf\u89c9\u5230\uff0c\u5e76\u4ea7\u751f\u4e0d\u9002\u3002 3. \u8bed\u97f3\u5408\u6210\uff1a\u5f53\u63d2\u5165\u6216\u4fee\u6539\u6587\u672c\u65f6\uff0c\u9700\u8981\u8bed\u97f3\u5408\u6210\u6a21\u5757\u6765\u751f\u6210\u65b0\u7684\u58f0\u97f3\u3002\u8bed\u97f3\u5408\u6210\u7684\u6700\u5927\u6311\u6218\u662f\u81ea\u7136\u548c\u6d41\u7545\uff0c\u5bf9\u4e8e\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u5c24\u5176\u91cd\u8981\uff0c\u56e0\u4e3a\u5982\u679c\u53ea\u4fee\u6539\u8bed\u97f3\u7684\u4e00\u90e8\u5206\uff0c\u7a0d\u6709\u4e0d\u8fde\u8d2f\u5c31\u4f1a\u975e\u5e38\u660e\u663e\u3002\u800c\u4e14\u7814\u7a76\u5458\u4eec\u671f\u671b\u5728\u4f7f\u7528\u8bed\u97f3\u7f16\u8f91\u6280\u672f\u7684\u65f6\u5019\uff0c\u53ef\u4ee5\u968f\u610f\u8fdb\u884c\u7f16\u8f91\u800c\u4e0d\u9700\u8981\u51c6\u5907\u8db3\u91cf\u7684\u8bed\u97f3\u6570\u636e\u53bb\u5fae\u8c03\u6a21\u578b\u3002\u56e0\u6b64\u4e00\u4e2a\u96f6\u6837\u672c\u4e0a\u4e0b\u6587\u611f\u77e5\uff08zero-shot context-aware\uff09\u7684 TTS \u662f\u5fc5\u4e0d\u53ef\u5c11\u7684\u3002 4. \u586b\u5145\u8bcd\u68c0\u6d4b\uff1a\u586b\u5145\u8bcd\u68c0\u6d4b\u6a21\u5757\u53ef\u4ee5\u81ea\u52a8\u68c0\u6d4b\u8bed\u97f3\u4e2d\u7684\u586b\u5145\u8bcd\uff0c\u7528\u6237\u53ef\u4ee5\u9009\u62e9\u624b\u52a8\u5220\u9664\u90e8\u5206\u6216\u81ea\u52a8\u5168\u90e8\u5220\u9664\u3002\u4e0a\u6587\u63d0\u5230\uff0c\u90e8\u5206 ASR \u7cfb\u7edf\u5e76\u4e0d\u80fd\u68c0\u6d4b\u5230\u5168\u90e8\u586b\u5145\u8bcd\uff0c\u6240\u4ee5\u6b64\u65f6\u5c31\u6ca1\u529e\u6cd5\u901a\u8fc7\u6587\u672c\u5220\u9664\u586b\u5145\u8bcd\u6765\u7f16\u8f91\u8bed\u97f3\u3002\u6709\u7684\u8bcd\u662f\u5426\u662f\u586b\u5145\u8bcd\u53ef\u80fd\u53d6\u51b3\u4e8e\u8bed\u5883\uff0c\u6bd4\u5982\u201cyou know\u201d\u662f\u82f1\u8bed\u4e2d\u5e38\u7528\u7684\u586b\u5145\u8bcd\uff0c\u4f46\u662f\u5728\u201cDo you know him?\u201d\u8fd9\u53e5\u8bdd\u4e2d\u5b83\u5e76\u4e0d\u662f\u586b\u5145\u8bcd\uff0c\u8fd9\u65f6\u5c31\u9700\u8981\u4e00\u4e2a\u8bed\u8a00\u6a21\u578b\u6765\u8fdb\u884c\u5224\u65ad\u3002 \u7814\u7a76\u5458\u4eec\u9996\u5148\u8c03\u7528\u5fae\u8f6f\u4e91\u8ba1\u7b97\u5e73\u53f0 Azure \u4e0a\u7684 ASR \u670d\u52a1\u5c06\u4e0a\u4f20\u7684\u8bed\u97f3\u6587\u4ef6\u8f6c\u5316\u4e3a\u6587\u672c\uff0c\u540c\u65f6\u8c03\u7528\u81ea\u884c\u7814\u53d1\u7684\u586b\u5145\u8bcd\u68c0\u6d4b\u6a21\u578b\uff0c\u5e76\u5c06\u586b\u5145\u8bcd\u68c0\u6d4b\u7ed3\u679c\u548c ASR \u8bc6\u522b\u7ed3\u679c\u5408\u5e76\u3002\u7136\u540e\u5c31\u53ef\u4ee5\u5bf9\u6587\u672c\u8fdb\u884c\u7f16\u8f91\u2014\u2014\u5bf9\u4e8e\u5220\u9664\u64cd\u4f5c\uff0c\u7cfb\u7edf\u4f1a\u6839\u636e\u5bf9\u9f50\u7ed3\u679c\u5220\u9664\u5bf9\u5e94\u7684\u8bed\u97f3\u7247\u6bb5\uff1b\u5bf9\u4e8e\u63d2\u5165\u64cd\u4f5c\uff0c\u7cfb\u7edf\u4f1a\u8c03\u7528\u8bed\u97f3\u5408\u6210\u6a21\u578b\u5408\u6210\u8981\u63d2\u5165\u7684\u8bed\u97f3\u5e76\u63d2\u5165\u539f\u6709\u8bed\u97f3\u4e2d\u3002\u4e0b\u9762\u662f\u51e0\u4e2a\u901a\u8fc7\u4e0a\u8ff0\u65b9\u6cd5\u5b8c\u6210\u7684\u57fa\u4e8e\u6587\u672c\u7684\u8bed\u97f3\u7f16\u8f91\u6837\u4f8b\uff1a \u4fee\u6539\u6587\u672c\u6837\u4f8b \u539f\u59cb\u6587\u672c\uff1aunderstand for the question and answer benchmark we&#8217;re also the first reach human parity \u4fee\u6539\u540e\u6587\u672c\uff1aunderstand for [&hellip;]<\/p>\n","protected":false},"author":34512,"featured_media":1070595,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-content-parent":1012650,"msr_hide_image_in_river":0,"footnotes":""},"research-area":[],"msr-locale":[268881],"msr-post-option":[],"class_list":["post-1069872","msr-blog-post","type-msr-blog-post","status-publish","has-post-thumbnail","hentry","msr-locale-zh_cn"],"msr_assoc_parent":{"id":1012650,"type":"lab"},"_links":{"self":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1069872","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post"}],"about":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-blog-post"}],"author":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/users\/34512"}],"version-history":[{"count":3,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1069872\/revisions"}],"predecessor-version":[{"id":1080006,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1069872\/revisions\/1080006"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media\/1070595"}],"wp:attachment":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1069872"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1069872"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1069872"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1069872"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}