{"id":6944,"date":"2024-09-07T19:06:50","date_gmt":"2024-09-07T11:06:50","guid":{"rendered":"http:\/\/loc.jiyun.net\/?p=6708"},"modified":"2024-09-07T19:06:50","modified_gmt":"2024-09-07T11:06:50","slug":"%e9%98%bf%e9%87%8c%e5%bc%80%e6%ba%90%e6%95%b0%e5%ad%97%e4%ba%ba%e5%b7%a5%e5%85%b7%ef%bc%9a%e8%ae%a9%e7%85%a7%e7%89%87%e5%bc%80%e5%8f%a3%e8%af%b4%e8%af%9d%ef%bc%8c%e6%95%88%e6%9e%9c%e6%83%8a%e8%89%b3","status":"publish","type":"post","link":"http:\/\/puo.cn\/?p=6944","title":{"rendered":"\u963f\u91cc\u5f00\u6e90\u6570\u5b57\u4eba\u5de5\u5177\uff1a\u8ba9\u7167\u7247\u5f00\u53e3\u8bf4\u8bdd\uff0c\u6548\u679c\u60ca\u8273\u5230\u7206\uff01"},"content":{"rendered":"
\n

\u60f3\u8c61\u4e00\u4e0b\uff0c\u5f53\u4f60\u5728\u535a\u7269\u9986\u4e2d\u6b23\u8d4f\u4e00\u5e45\u53e4\u8001\u7684\u4eba\u7269\u753b\u4f5c\u65f6\uff0c \u4f60\u4e0d\u518d\u9700\u8981\u9605\u8bfb\u65c1\u8fb9\u67af\u71e5\u7684\u8bf4\u660e\u6587\u5b57\u3002\u753b\u4f5c\u672c\u8eab\u5c31\u4f1a\u5411\u4f60\u8bb2\u8ff0\u5b83\u7684\u6545\u4e8b\uff0c\u63cf\u8ff0\u5b83\u7684\u521b\u4f5c\u80cc\u666f\uff0c\u751a\u81f3\u900f\u9732\u51fa\u753b\u5bb6\u7684\u60c5\u611f\u548c\u610f\u56fe\u3002\u8fd9\u4e0d\u4ec5\u662f\u4e00\u6b21\u89c6\u89c9\u7684\u4eab\u53d7\uff0c\u66f4\u662f\u4e00\u6b21\u5fc3\u7075\u7684\u89e6\u52a8\u3002<\/p>\n

\u751a\u81f3\u5728\u6211\u4eec\u7684\u65e5\u5e38\u751f\u6d3b\u4e2d\uff0c\u5f53\u6211\u4eec\u7ffb\u770b\u5bb6\u5ead\u76f8\u518c\u65f6\uff0c\u90a3\u4e9b\u5c18\u5c01\u7684\u8bb0\u5fc6\u53ef\u4ee5\u88ab\u91cd\u65b0\u5524\u9192\u3002\u6bcf\u4e00\u5f20\u7167\u7247\u90fd\u80fd\u8bb2\u8ff0\u5b83\u80cc\u540e\u7684\u6545\u4e8b\uff0c\u8ba9\u6211\u4eec\u91cd\u65b0\u4f53\u9a8c\u90a3\u4e9b\u73cd\u8d35\u7684\u65f6\u523b\u3002<\/p>\n

\u672c\u6587\u6211\u5c06\u4ecb\u7ecd\u963f\u91cc\u8682\u8681\u96c6\u56e2\u5f00\u6e90\u7684 \u201c\u8ba9\u7167\u7247\u8bf4\u8bdd\u201d \u7684 AI \u5f00\u6e90\u9879\u76ee \u2014\u2014\u00a0EchoMimic\u3002<\/p>\n

<\/figure>\n

EchoMimic \u662f\u4e00\u4e2a\u521b\u65b0\u7684\u8096\u50cf\u89c6\u9891\u52a8\u753b\u751f\u6210\u6280\u672f\uff0c\u5b83\u80fd\u591f\u901a\u8fc7\u97f3\u9891\u4fe1\u53f7\u548c\u9762\u90e8\u5173\u952e\u70b9\u4e24\u79cd\u9a71\u52a8\u65b9\u5f0f\uff0c\u4ee5\u53ca\u5b83\u4eec\u7684\u7ec4\u5408\u6765\u751f\u6210\u751f\u52a8\u7684\u8096\u50cf\u89c6\u9891\u3002\u8be5\u6280\u672f\u57fa\u4e8e Stable Diffusion\uff08SD\uff09\u6846\u67b6\uff0c\u91c7\u7528\u4e86 Latent Diffusion Model\uff08LDM\uff09\u548c Variational Autoencoder\uff08VAE\uff09\uff0c\u901a\u8fc7\u5411\u6f5c\u7a7a\u95f4\u8868\u793a\u4e2d\u5f15\u5165\u9ad8\u65af\u566a\u58f0\u5e76\u8fdb\u884c\u53bb\u566a\u8fc7\u7a0b\u6765\u751f\u6210\u56fe\u50cf\u3002<\/p>\n

EchoMimic \u7684\u6838\u5fc3\u662f Denoising U-Net \u67b6\u6784\uff0c\u5b83\u5305\u62ec Reference U-Net\u3001Landmark Encoder \u548c Audio Encoder \u4e09\u4e2a\u4e13\u95e8\u7684\u6a21\u5757\uff0c\u4ee5\u53ca Temporal Attention Layer\uff0c\u7528\u4e8e\u786e\u4fdd\u89c6\u9891\u5e8f\u5217\u7684\u65f6\u95f4\u4e00\u81f4\u6027\u3002<\/p>\n

<\/figure>\n

\u8fd1\u671f\u70ed\u6587<\/span><\/h2>\n