{"id":25807,"date":"2025-04-22T20:05:51","date_gmt":"2025-04-22T12:05:51","guid":{"rendered":"http:\/\/139.9.1.231\/?p=25807"},"modified":"2025-05-11T12:11:53","modified_gmt":"2025-05-11T04:11:53","slug":"dolphin-asr-model-llm","status":"publish","type":"post","link":"http:\/\/139.9.1.231\/index.php\/2025\/04\/22\/dolphin-asr-model-llm\/","title":{"rendered":"Dolphin -\u652f\u6301\u4e1c\u65b940\u8bed\u79cd+\u4e2d\u56fd22\u65b9\u8a00\u7684\u65b0SOTA\u8bed\u97f3\u5927\u6a21\u578b"},"content":{"rendered":"\n<ul class=\"has-light-pink-background-color has-background\"><li><em>\u8bba\u6587\u9898\u76ee\uff1a<strong>Dolphin: A Large-Scale Automatic Speech Recognition Model for Eastern Languages<\/strong><\/em><\/li><li><em>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/arxiv.org\/abs\/2503.20212\"><strong>https:\/\/arxiv.org\/abs\/2503.20212<\/strong><\/a><\/em><\/li><li><em><strong>Github\uff1a<a href=\"https:\/\/github.com\/DataoceanAI\/Dolphin\">https:\/\/github.com\/DataoceanAI\/Dolphin<\/a><\/strong><\/em><\/li><li><em>Huggingface\uff1a<a href=\"https:\/\/huggingface.co\/DataoceanAI\">https:\/\/huggingface.co\/DataoceanAI<\/a><\/em><\/li><li><em>Modelscope\uff1a<a href=\"https:\/\/www.modelscope.cn\/organization\/DataoceanAI\">https:\/\/www.modelscope.cn\/organization\/DataoceanAI<\/a><\/em><\/li><li><em>OpenI\u542f\u667a\u793e\u533a\uff1a<a href=\"https:\/\/openi.pcl.ac.cn\/DataoceanAI\/Dolphin\"><strong>https:\/\/openi.pcl.ac.cn\/DataoceanAI\/Dolphin<\/strong><\/a><\/em><\/li><li><em>\u652f\u6301\u7684\u8bed\u79cd\uff1a<a href=\"https:\/\/github.com\/DataoceanAI\/Dolphin\/blob\/main\/languages.md\"><strong>https:\/\/github.com\/DataoceanAI\/Dolphin\/blob\/main\/languages.md<\/strong><\/a><\/em><\/li><\/ul>\n\n\n\n\n\n<p>\u5728\u5f53\u4eca\u6570\u5b57\u5316\u65f6\u4ee3\uff0c\u8bed\u97f3\u8bc6\u522b\u6280\u672f\u5df2\u6210\u4e3a\u4eba\u673a\u4ea4\u4e92\u7684\u5173\u952e\u6865\u6881\uff0c\u5e7f\u6cdb\u5e94\u7528\u4e8e\u667a\u80fd\u5ba2\u670d\u3001\u8bed\u97f3\u52a9\u624b\u3001\u4f1a\u8bae\u8f6c\u5f55\u7b49\u4f17\u591a\u9886\u57df\u3002\u7136\u800c\uff0c\u5bf9\u4e8e\u4e1c\u65b9\u8bed\u8a00\u7684\u8bc6\u522b\u5982\u8d8a\u5357\u8bed\u3001\u7f05\u7538\u8bed\u7b49\uff0c\u73b0\u6709\u6a21\u578b\u5f80\u5f80\u8868\u73b0\u4e0d\u4f73\uff0c\u96be\u4ee5\u6ee1\u8db3\u7528\u6237\u7684\u9700\u6c42\u3002\u4e3a\u89e3\u51b3\u8fd9\u4e00\u96be\u9898\uff0c\u6d77\u5929\u745e\u58f0\u643a\u624b\u6e05\u534e\u5927\u5b66\u7535\u5b50\u5de5\u7a0b\u7cfb\u8bed\u97f3\u4e0e\u97f3\u9891\u6280\u672f\u5b9e\u9a8c\u5ba4\uff0c\u5171\u540c\u63a8\u51fa\u4e86Dolphin \u2014\u2014 <strong>\u4e00\u6b3e\u4e13\u4e3a\u4e1c\u65b9\u8bed\u8a00\u8bbe\u8ba1\u7684\u8bed\u97f3\u5927\u6a21\u578b<\/strong>\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" width=\"1024\" height=\"296\" src=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-21-1024x296.png\" alt=\"\" class=\"wp-image-25843\" srcset=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-21-1024x296.png 1024w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-21-300x87.png 300w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-21-768x222.png 768w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-21.png 1270w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><figcaption>Dolphin \u91c7\u7528\u7684\u591a\u4efb\u52a1\u683c\u5f0f\uff0c\u5176\u4e3b\u8981\u6cbf\u7528\u4e86 OpenAI Whisper\u7684<br>\u683c\u5f0f\u3002Dolphin \u4e13\u6ce8\u4e8e\u81ea\u52a8\u8bed\u97f3\u8bc6\u522b (ASR)\uff0c\u4e0d\u652f\u6301\u7ffb\u8bd1\u4efb\u52a1\u3002\u6b64\u5916\uff0cDolphin \u5f15\u5165\u4e86\u7279\u5b9a\u533a\u57df\u7684\u6807\u8bb0\uff0c\u4ece\u800c\u652f\u6301\u65b9\u8a00\u3002<\/figcaption><\/figure>\n\n\n\n<p>Dolphin \u662f\u7531 Dataocean AI \u4e0e\u6e05\u534e\u5927\u5b66\u5408\u4f5c\u5f00\u53d1\u7684\u591a\u8bed\u8a00\u3001\u591a\u4efb\u52a1 ASR \u6a21\u578b\u3002\u5b83<strong>\u652f\u6301\u4e1c\u4e9a\u3001\u5357\u4e9a\u3001\u4e1c\u5357\u4e9a\u548c\u4e2d\u4e1c\u5730\u533a\u7684 40 \u79cd\u4e1c\u65b9\u8bed\u8a00\uff0c\u540c\u65f6\u8fd8\u652f\u6301 22 \u79cd\u4e2d\u56fd\u65b9\u8a00<\/strong>\u3002\u8be5\u6a21\u578b\u57fa\u4e8e\u8d85\u8fc7<strong> 21 \u4e07\u5c0f\u65f6\u7684\u6570\u636e\u8fdb<\/strong>\u884c\u8bad\u7ec3\uff0c\u5176\u4e2d\u5305\u62ec DataoceanAI \u7684\u4e13\u6709\u6570\u636e\u96c6\u548c\u5f00\u6e90\u6570\u636e\u96c6\u3002\u8be5\u6a21\u578b\u53ef\u4ee5\u6267\u884c<strong>\u8bed\u97f3\u8bc6\u522b\u3001\u8bed\u97f3\u6d3b\u52a8\u68c0\u6d4b (VAD)\u3001\u8bed\u97f3\u5206\u5272\u548c\u8bed\u8a00\u8bc6\u522b (LID)<\/strong>\u3002<\/p>\n\n\n\n<h3><strong>\u4e8c\u3001\u521b\u65b0\u6280\u672f\u67b6\u6784&nbsp;<\/strong><\/h3>\n\n\n\n<ul><li>\u6a21\u578b\u7ed3\u6784&nbsp; &nbsp;&nbsp;<\/li><\/ul>\n\n\n\n<p>Dolphin\u7f51\u7edc\u7ed3\u6784\u57fa\u4e8eCTC-Attention\u67b6\u6784\uff0cE-Branchformer\u7f16\u7801\u5668\u548cTransformer\u89e3\u7801\u5668\uff0c\u5e76\u5f15\u5165\u4e864\u500d\u4e0b\u91c7\u6837\u5c42\uff0c\u4ee5\u5b9e\u73b0\u9ad8\u6548\u7684\u5927\u89c4\u6a21\u591a\u8bed\u8a00\u8bed\u97f3\u8bc6\u522b\u6a21\u578b\u7684\u8bad\u7ec3\u3002CTC-Attention\u67b6\u6784\u7ed3\u5408\u4e86CTC\u7684\u5e8f\u5217\u5efa\u6a21\u80fd\u529b\u548c\u6ce8\u610f\u529b\u673a\u5236\u7684\u4e0a\u4e0b\u6587\u6355\u6349\u80fd\u529b\uff0c\u80fd\u591f\u6709\u6548\u63d0\u5347\u6a21\u578b\u7684\u8bc6\u522b\u51c6\u786e\u6027\u548c\u6548\u7387\u3002E-Branchformer\u7f16\u7801\u5668\u91c7\u7528\u5e76\u884c\u5206\u652f\u7ed3\u6784\uff0c\u80fd\u591f\u66f4\u6709\u6548\u5730\u6355\u6349\u8f93\u5165\u8bed\u97f3\u4fe1\u53f7\u7684\u5c40\u90e8\u548c\u5168\u5c40\u4f9d\u8d56\u5173\u7cfb\uff0c\u4e3a\u6a21\u578b\u63d0\u4f9b\u4e86\u66f4\u4e30\u5bcc\u7684\u7279\u5f81\u8868\u793a\u3002\u89e3\u7801\u5668\u90e8\u5206\u5219\u91c7\u7528\u4e86\u5728\u5e8f\u5217\u5230\u5e8f\u5217\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u8272\u7684Transformer\uff0c\u80fd\u591f\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u6587\u672c\u8f93\u51fa\u3002\u4e3a\u4e86\u8fdb\u4e00\u6b65\u63d0\u9ad8\u8bad\u7ec3\u6548\u7387\u548c\u6027\u80fd\uff0c\u6211\u4eec\u5728\u6a21\u578b\u4e2d\u5f15\u5165\u4e864\u500d\u4e0b\u91c7\u6837\u5c42\u3002\u8fd9\u4e00\u5c42\u53ef\u4ee5\u51cf\u5c11\u8f93\u5165\u7279\u5f81\u7684\u5e8f\u5217\u957f\u5ea6\uff0c\u4ece\u800c\u52a0\u901f\u8ba1\u7b97\u8fc7\u7a0b\uff0c\u540c\u65f6\u4fdd\u7559\u5173\u952e\u7684\u8bed\u97f3\u4fe1\u606f\uff0c\u786e\u4fdd\u6a21\u578b\u7684\u8bc6\u522b\u6548\u679c\u4e0d\u53d7\u5f71\u54cd\u3002<\/p>\n\n\n\n<ul><li>\u591a\u4efb\u52a1\u683c\u5f0f<\/li><\/ul>\n\n\n\n<p>Dolphin \u501f\u9274\u4e86 Whisper \u548c OWSM \u7684\u521b\u65b0\u8bbe\u8ba1\u65b9\u6cd5\uff0c\u4f46\u4e13\u6ce8\u4e8eASR \u8fdb\u884c\u4e86\u82e5\u5e72\u5173\u952e\u4fee\u6539\u3002<strong>Dolphin \u4e0d\u652f\u6301\u7ffb\u8bd1\u4efb\u52a1\uff0c\u5e76\u4e14\u53bb\u6389\u4e86previous text\u53ca\u5176\u76f8\u5173\u6807\u8bb0\u7684\u4f7f\u7528\uff0c\u8fd9\u7b80\u5316\u4e86\u8f93\u5165\u683c\u5f0f\u5e76\u51cf\u5c11\u4e86\u6f5c\u5728\u7684\u590d\u6742\u6027<\/strong>\u3002<strong>Dolphin\u5f15\u5165\u4e86\u4e24\u7ea7\u8bed\u79cd\u6807\u7b7e\u7cfb\u7edf<\/strong>\uff0c\u4ee5\u4fbf\u66f4\u597d\u5730\u5904\u7406\u8bed\u8a00\u548c\u5730\u533a\u7684\u591a\u6837\u6027\u3002\u7b2c\u4e00\u4e2a\u6807\u7b7e\u6307\u5b9a\u8bed\u79cd\uff08\u4f8b\u5982:\u00a0&lt;zh>\u00a0\u3001\u00a0&lt;ja>\uff09\uff0c\u7b2c\u4e8c\u4e2a\u6807\u7b7e\u6307\u5b9a\u5730\u533a\uff08\u4f8b\u5982\u00a0&lt;CN>\u00a0\u3001\u00a0&lt;JP>\uff09\u3002\u00a0\u6bd4\u5982\uff1a<strong><code>&lt;ru>&lt;RU><\/code>\u00a0\u8868\u793a\u4fc4\u7f57\u65af\u7684\u4fc4\u8bed\uff0c\u800c\u00a0<code>&lt;ru>&lt;BY><\/code>\u00a0\u8868\u793a\u767d\u4fc4\u7f57\u65af\u7684\u4fc4\u8bed<\/strong>\u3002\u8fd9\u79cd<strong>\u5206\u5c42\u65b9\u6cd5\u4f7f\u6a21\u578b\u80fd\u591f\u6355\u6349\u540c\u4e00\u79cd\u8bed\u8a00\u5185\u4e0d\u540c\u65b9\u8a00\u548c\u53e3\u97f3\u4e4b\u95f4\u7684\u5dee\u5f02<\/strong>\uff0c\u4ee5\u53ca\u540c\u4e00\u5730\u533a\u5185\u4e0d\u540c\u8bed\u8a00\u4e4b\u95f4\u7684\u76f8\u4f3c\u6027\uff0c\u4ece\u800c\u63d0\u9ad8\u4e86\u6a21\u578b\u533a\u5206\u5bc6\u5207\u76f8\u5173\u7684\u65b9\u8a00\u7684\u80fd\u529b\uff0c\u5e76\u901a\u8fc7\u5728\u8bed\u8a00\u548c\u5730\u533a\u4e4b\u95f4\u5efa\u7acb\u8054\u7cfb\u589e\u5f3a\u4e86\u5176\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n\n\n\n<h3><strong>\u4e09\u3001\u5f3a\u5927\u7684\u6570\u636e\u57fa\u7840&nbsp;<\/strong><\/h3>\n\n\n\n<p>Dolphin\u7684\u8bad\u7ec3\u6570\u636e\u96c6\u6574\u5408\u4e86\u6d77\u5929\u745e\u58f0\u3010Dataocean AI\u3011\u7684\u4e13\u6709\u6570\u636e\u548c\u591a\u4e2a\u5f00\u6e90\u6570\u636e\u96c6\uff0c\u603b\u65f6\u957f\u8d85\u8fc720\u4e07\u5c0f\u65f6\uff0c\u6db5\u76d640\u4e2a\u4e1c\u65b9\u8bed\u79cd\u3002\u5176\u4e2d\uff0c\u6d77\u5929\u745e\u58f0\u6570\u636e\u96c6\u5305\u542b137,712\u5c0f\u65f6\u7684\u97f3\u9891\uff0c\u8986\u76d638\u4e2a\u4e1c\u65b9\u8bed\u79cd\u3002\u8fd9\u4e9b\u9ad8\u8d28\u91cf\u3001\u591a\u6837\u5316\u7684\u6570\u636e\u4e3a\u6a21\u578b\u7684\u8bad\u7ec3\u63d0\u4f9b\u4e86\u575a\u5b9e\u7684\u57fa\u7840\uff0c\u4f7f\u5176\u80fd\u591f\u66f4\u597d\u5730\u9002\u5e94\u4e0d\u540c\u8bed\u8a00\u548c\u65b9\u8a00\u7684\u8bed\u97f3\u7279\u5f81\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><img loading=\"lazy\" src=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-25.png\" alt=\"\" class=\"wp-image-25896\" width=\"554\" height=\"298\" srcset=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-25.png 946w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-25-300x162.png 300w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-25-768x415.png 768w\" sizes=\"(max-width: 554px) 100vw, 554px\" \/><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" src=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-26-1024x491.png\" alt=\"\" class=\"wp-image-25899\" width=\"694\" height=\"332\" srcset=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-26-1024x491.png 1024w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-26-300x144.png 300w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-26-768x368.png 768w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-26.png 1453w\" sizes=\"(max-width: 694px) 100vw, 694px\" \/><figcaption>\u6e05\u7406\u540e\u6570\u636e\u96c6\u4e2d 40 \u79cd\u4e1c\u65b9\u8bed\u8a00\u7684\u6570\u636e\u65f6\u957f\u5206\u5e03\uff08\u4ee5\u5bf9\u6570\u523b\u5ea6\u8868\u793a\uff09\u3002\u5176\u4e2d 36 \u79cd\u8bed\u8a00\u7684\u6570\u636e\u65f6\u957f\u8d85\u8fc7 100 \u5c0f\u65f6\uff0c16 \u79cd\u8bed\u8a00\u7684\u6570\u636e\u65f6\u957f\u8d85\u8fc7 1000 \u5c0f\u65f6\u3002<\/figcaption><\/figure>\n\n\n\n<p>\u6570\u636e\u5904\u7406\uff1a\u5bf9\u4e8e\u50cf YODAS \u8fd9\u6837\u5305\u542b\u4eba\u5de5\u6ce8\u91ca\u548c ASR \u751f\u6210\u7684\u8f6c\u5f55\u672c\u7684\u6570\u636e\u96c6\uff0c\u6211\u4eec\u53ea\u4f7f\u7528\u4eba\u5de5\u6ce8\u91ca\u7684\u90e8\u5206\u3002\u56e0\u6b64\uff0c\u6211\u4eec\u7684\u5927\u90e8\u5206\u8bad\u7ec3\u6570\u636e\u90fd\u662f\u624b\u52a8\u8f6c\u5f55\u7684\uff0c\u4ee5\u786e\u4fdd\u66f4\u9ad8\u7684\u8f6c\u5f55\u8d28\u91cf\u3002\u8fd9\u79cd\u6570\u636e\u8d28\u91cf\uff0c\u5c24\u5176\u662f\u8f6c\u5f55\u672c\u7684\u8d28\u91cf\uff0c\u662f\u4f7f\u6a21\u578b\u5373\u4f7f\u5728\u6a21\u578b\u89c4\u6a21\u8f83\u5c0f\u7684\u60c5\u51b5\u4e0b\u4e5f\u80fd\u5b9e\u73b0\u663e\u8457\u4f18\u4e8e Whisper \u8bc6\u522b\u6027\u80fd\u7684\u5173\u952e\u56e0\u7d20\u3002\u5bf9\u4e8e<strong>\u65f6\u95f4\u6233<\/strong>\uff0c\u91c7\u7528\u4e0e Whisper \u76f8\u540c\u7684\u53e5\u5b50\u7ea7\u65f6\u95f4\u6233\u65b9\u6cd5\uff0c\u5176\u4e2d\u65f6\u95f4\u6233\u6807\u8bb0\u6807\u8bb0\u6bcf\u4e2a\u53e5\u5b50\u7684\u8d77\u59cb\u548c\u7ed3\u675f\u3002\u5bf9\u4e8e\u957f\u97f3\u9891\u5f55\u97f3\uff08\u901a\u5e38\u957f\u8fbe\u51e0\u5206\u949f\uff09\uff0c\u4f1a\u5728\u6570\u636e\u9884\u5904\u7406\u8fc7\u7a0b\u4e2d\u5c06\u5176\u5206\u5272\u6210\u8f83\u5c0f\u7684\u7247\u6bb5\uff0c\u7136\u540e\u5c06\u5b83\u4eec\u5408\u5e76\u4e3a\u957f\u97f3\u9891\u5e8f\u5217\u3002<\/p>\n\n\n\n<p><strong>\u8bad\u7ec3\u4f18\u5316\uff1a<\/strong><\/p>\n\n\n\n<p>\u5728\u8bad\u7ec3\u6570\u636e\u7684\u521d\u59cb\u7248\u672c\u4e2d\uff0c\u6211\u4eec\u76f4\u63a5\u4f7f\u7528\u4e86\u6e05\u7406\u540e\u7684\u6570\u636e\u96c6\u3002\u7136\u800c\uff0c\u4e00\u4e2a\u4e3b\u8981\u95ee\u9898\u662f\u77ed\u97f3\u9891\u6837\u672c\u7684\u6bd4\u4f8b\u8fc7\u9ad8\u3002\u5927\u591a\u6570\u97f3\u9891\u7247\u6bb5\u7684\u65f6\u957f\u7ea6\u4e3a 5 \u79d2\uff0c\u5bfc\u81f4\u8de8\u591a\u79cd\u8bed\u8a00\u7684\u5220\u9664\u9519\u8bef\u7387\u8fc7\u9ad8\u3002\u8fd9\u4e2a\u95ee\u9898\u4e0e\u5927\u591a\u6570\u8bad\u7ec3\u6570\u636e\u7531\u77ed\u97f3\u9891\u6837\u672c\u7ec4\u6210\u8fd9\u4e00\u4e8b\u5b9e\u76f8\u7b26\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u5c1d\u8bd5\u4e86\u4e00\u79cd\u66ff\u4ee3\u65b9\u6cd5\uff0c<strong>\u5c06\u6e05\u7406\u540e\u7684\u97f3\u9891\u6570\u636e\u8fde\u63a5\u6210 25-30 \u79d2\u7684\u957f\u7247\u6bb5\u3002<\/strong>\u8fd9\u663e\u8457\u964d\u4f4e\u4e86\u8f83\u9ad8\u7684\u5220\u9664\u9519\u8bef\u7387\u3002<strong>\u867d\u7136\u8fd9\u79cd\u65b9\u6cd5\u5bfc\u81f4\u63d2\u5165\u9519\u8bef\u7387\u7565\u6709\u589e\u52a0\uff0c\u4f46\u6574\u4f53\u8bc6\u522b\u6027\u80fd\u6709\u6240\u63d0\u5347<\/strong>\uff0c\u5e73\u5747\u5b57\u8bcd\u9519\u8bef\u7387 (WER) \u964d\u4f4e\u4e86 9.01%\u3002<\/p>\n\n\n\n<h3><strong>\u56db\u3001\u5353\u8d8a\u6027\u80fd\u8868\u73b0&nbsp;<\/strong><\/h3>\n\n\n\n<p>\u901a\u8fc7\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u67b6\u6784\u548c\u5927\u89c4\u6a21\u7684\u8bad\u7ec3\u6570\u636e\uff0cDolphin\u5728\u591a\u79cd\u8bed\u8a00\u4e0a\u7684\u8bcd\u9519\u8bef\u7387\uff08WER\uff09\u663e\u8457\u4f4e\u4e8e\u73b0\u6709\u5f00\u6e90\u6a21\u578b\u3002<\/p>\n\n\n\n<p>\u4f8b\u5982\uff0c\u5728\u6d77\u5929\u745e\u58f0\u6570\u636e\u96c6\u4e0a\uff0cDolphin \u6a21\u578b\u7684\u5e73\u5747WER\u4e3a31.5%\uff0csmall\u6a21\u578b\u4e3a24.5%\uff0cmedium\u6a21\u578b\u4e3a22.2%\uff1b\u5728CommonVoice\u6570\u636e\u96c6\u4e0a\uff0cDolphin \u6a21\u578b\u7684\u5e73\u5747WER\u4e3a37.2%\uff0csmall\u6a21\u578b\u4e3a27.4%\uff0cmedium\u6a21\u578b\u4e3a25.0%\u3002\u5373\u4f7f\u4e0eWhisper large-v3\u6a21\u578b\u76f8\u6bd4\uff0cDolphin\u5728\u6a21\u578b\u89c4\u6a21\u66f4\u5c0f\u7684\u60c5\u51b5\u4e0b\uff0c\u6027\u80fd\u4e5f\u66f4\u4e3a\u51fa\u8272\u3002\u4ee5\u4e2d\u6587\u4e3a\u4f8b\uff0cDolphin\u4e2d\u6a21\u578b\u7684WER\u4ec5\u4e3a9.2%\uff0c\u800cWhisper large-v3\u6a21\u578b\u4e3a27.9%\u3002&nbsp;\u5728KeSpeech \uff08\u5305\u542b\u4e00\u4e2a\u666e\u901a\u8bdd\u5b50\u96c6\u548c\u516b\u4e2a\u4e2d\u56fd\u65b9\u8a00\u5b50\u96c6\uff09\u6d4b\u8bd5\u96c6\u4e0a\uff0cDolphin\u6a21\u578b\u8868\u73b0\u51fa\u4e86\u5353\u8d8a\u7684\u6548\u679c.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" width=\"627\" height=\"804\" src=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-24.png\" alt=\"\" class=\"wp-image-25891\" srcset=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-24.png 627w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-24-234x300.png 234w\" sizes=\"(max-width: 627px) 100vw, 627px\" \/><\/figure>\n\n\n\n<h3><strong>\u4e94\u3001\u6280\u672f\u6311\u6218<\/strong><\/h3>\n\n\n\n<p><strong>\u5185\u5b58\u5360\u7528\u95ee\u9898<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><img loading=\"lazy\" src=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-27.png\" alt=\"\" class=\"wp-image-25920\" width=\"546\" height=\"229\" srcset=\"http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-27.png 700w, http:\/\/139.9.1.231\/wp-content\/uploads\/2025\/04\/image-27-300x126.png 300w\" sizes=\"(max-width: 546px) 100vw, 546px\" \/><figcaption>\u56fe 3\uff1a&nbsp;\u6570\u636e\u52a0\u8f7d\u7b56\u7565\u4f18\u5316\u3002\u5047\u8bbe\u4e00\u4e2a\u8282\u70b9\u6709 4 \u4e2a GPU\uff0c\u6bcf\u4e2a GPU \u5206\u914d\u4e00\u4e2a\u5bf9\u5e94\u7684\u8fdb\u7a0b\uff0c\u79f0\u4e3a rank\u3002\u4f18\u5316\u524d\uff0c\u6bcf\u4e2a rank \u52a0\u8f7d\u6570\u636e\u96c6\u7684\u5b8c\u6574\u526f\u672c\uff0c\u8bb0\u4e3a {D0,D1,D2,D3}\u3002\u4f18\u5316\u540e\uff0c\u6bcf\u4e2a rank \u4ec5\u5206\u914d\u5176\u8ba1\u7b97\u6240\u9700\u7684\u6570\u636e\u96c6\u5b50\u96c6\u3002<\/figcaption><\/figure>\n\n\n\n<p>\u6211\u4eec\u7684\u8bad\u7ec3\u96c6\u5305\u542b 1.6 \u4ebf\u6761\u8bdd\u8bed\uff0c\u5728\u6570\u636e\u5904\u7406\u9636\u6bb5\u9047\u5230\u4e86\u5185\u5b58\u4e0d\u8db3 (OOM) \u95ee\u9898\u3002\u6211\u4eec\u5bf9\u6570\u636e\u5904\u7406\u7684 sampler\u3001dataset\u3001dataloader \u6a21\u5757\u8fdb\u884c\u4e86\u6df1\u5165\u5206\u6790\uff0c\u53d1\u73b0\u5927\u91cf\u7684 utterances \u5bfc\u81f4\u4e86\u5185\u5b58\u6ea2\u51fa\u3002PyTorch \u652f\u6301\u4e24\u79cd\u7c7b\u578b\u7684\u6570\u636e\u96c6\uff1amap-style \u548c iterable-style\u3002ESPnet&nbsp;\u4f7f\u7528\u7684\u662f map-style\u3002map-style \u6570\u636e\u96c6\u5c06 utterance \u7684\u5143\u6570\u636e\uff08utterance id \u4e0e\u6587\u672c\u3001\u97f3\u9891\u7684\u6620\u5c04\uff09\u52a0\u8f7d\u5230\u5185\u5b58\u4e2d\uff0c\u5185\u5b58\u5360\u7528\u968f\u7740\u8bad\u7ec3\u6570\u636e utterances \u7684\u6570\u91cf\u7ebf\u6027\u589e\u957f\u3002\u4e3a\u4e86\u63d0\u9ad8\u6570\u636e\u52a0\u8f7d\u901f\u5ea6\uff0cdataloader \u5185\u90e8\u4f1a\u6709\u591a\u4e2a worker \u8fdb\u884c\u6570\u636e\u9884\u53d6\uff0c\u8fd9\u8fdb\u4e00\u6b65\u589e\u52a0\u4e86\u7269\u7406\u673a\u7684\u5185\u5b58\u5360\u7528\uff0c\u6700\u7ec8\u5bfc\u81f4 OOM\u3002<\/p>\n\n\n\n<p>\u53d7 Zero-DP\u7684\u542f\u53d1\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u56fe\u00a03\u00a0\u4e2d\u7684<strong>\u6570\u636e\u5206\u7247\u7b56\u7565\u3002\u6211\u4eec\u4e0d\u518d\u52a0\u8f7d\u6574\u4e2a\u6570\u636e\u96c6\u526f\u672c\uff0c\u800c\u662f\u4f18\u5316\u6bcf\u4e2a Rank\uff0c\u4f7f\u5176\u4ec5\u52a0\u8f7d\u6570\u636e\u96c6\u4e2d\u5fc5\u8981\u7684\u5b50\u96c6<\/strong>\u3002\u8fd9\u79cd\u65b9\u6cd5\u663e\u8457\u51cf\u5c11\u4e86\u6bcf\u4e2a Rank \u7684\u5185\u5b58\u5360\u7528\uff0c\u4ece\u800c\u964d\u4f4e\u4e86\u7269\u7406\u673a\u4e0a\u7684\u6574\u4f53\u5185\u5b58\u6d88\u8017\u3002\u6b64\u5916\uff0c\u968f\u7740\u6570\u636e\u5e76\u884c\u5ea6\u7684\u63d0\u9ad8\uff0c\u5355\u4e2a\u8282\u70b9\u7684\u5185\u5b58\u5360\u7528\u5448\u7ebf\u6027\u4e0b\u964d\u3002<\/p>\n\n\n\n<p><strong>\u8bad\u7ec3\u6548\u7387\uff1a<\/strong><\/p>\n\n\n\n<p>\u5c06\u77ed\u97f3\u9891\u5408\u5e76\u6210\u957f\u97f3\u9891\u53ef\u4ee5\u663e\u8457\u63d0\u9ad8 GPU \u7684\u8ba1\u7b97\u5bc6\u5ea6\u548c\u5229\u7528\u7387\uff0c\u4ece\u800c\u663e\u8457\u63d0\u9ad8\u8bad\u7ec3\u6548\u7387\u3002\u5728\u6211\u4eec\u7684\u6570\u636e\u96c6\u4e2d\uff0c\u97f3\u9891\u65f6\u957f\u5448\u73b0\u51fa\u660e\u663e\u7684\u5de6\u504f\u5206\u5e03\uff0c\u77ed\u97f3\u9891\uff081-10 \u79d2\uff09\u5360\u6bd4\u8f83\u9ad8\uff0c\u957f\u97f3\u9891\uff0811-30 \u79d2\uff09\u5360\u6bd4\u8f83\u4f4e\u3002\u4e3a\u4e86\u4f7f\u97f3\u9891\u65f6\u957f\u5206\u5e03\u66f4\u52a0\u5747\u8861\uff0c<strong>\u6211\u4eec\u5c06\u77ed\u97f3\u9891\u5408\u5e76\uff0c\u5e76\u5c06\u5b83\u4eec\u5747\u5300\u5730\u91cd\u65b0\u5206\u914d\u5230 0-30 \u79d2\u8303\u56f4\u5185\u4ee5 5 \u79d2\u4e3a\u95f4\u9694\u7684\u6876\u4e2d\u3002<\/strong><\/p>\n\n\n\n<p>\u5728\u5904\u7406 21 \u4e07\u5c0f\u65f6\u7684\u5927\u89c4\u6a21\u6570\u636e\u96c6\u65f6\uff0c\u4f7f\u7528 ffmpeg \u5c06\u591a\u4e2a\u77ed\u97f3\u9891\u7269\u7406\u5408\u5e76\u6210\u957f\u97f3\u9891\u4f1a\u975e\u5e38\u8017\u65f6\u3002\u4e3a\u6b64\uff0c\u6211\u4eec\u91c7\u7528\u4e86\u66f4\u9ad8\u6548\u7684\u903b\u8f91\u5408\u5e76\u7b56\u7565\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u5728\u6570\u636e\u51c6\u5907\u9636\u6bb5\uff0c\u6211\u4eec\u4f7f\u7528\u5b57\u5178\u6765\u8868\u793a\u97f3\u9891\u5408\u5e76\u524d\u540e\u7684\u6620\u5c04\u5173\u7cfb\uff0c\u5e76\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u52a8\u6001\u5730\u5408\u5e76\u97f3\u9891\u3002<\/p>\n\n\n\n<p>\u901a\u8fc7\u4f18\u5316\u5408\u5e76\u7b56\u7565\uff0c\u5c0f\u6a21\u578b\u5355\u6b21 epoch \u8bad\u7ec3\u65f6\u95f4\u4ece 64 \u5c0f\u65f6\u5927\u5e45\u7f29\u77ed\u81f3 28.6 \u5c0f\u65f6\uff0c\u8bad\u7ec3\u901f\u5ea6\u63d0\u5347 123.78%\uff0c\u5927\u5927\u52a0\u901f\u4e86\u6a21\u578b\u8fed\u4ee3\u8fdb\u7a0b\u3002<\/p>\n\n\n\n<h3><strong>\u516d\u3001\u5f00\u6e90\u4e0e\u793e\u533a\u8d21\u732e&nbsp;<\/strong><\/h3>\n\n\n\n<p>      \u4e3a\u4fc3\u8fdb\u8bed\u97f3\u8bc6\u522b\u6280\u672f\u7684\u8fdb\u4e00\u6b65\u53d1\u5c55\uff0cDolphin\u7684\u8bad\u7ec3\u6a21\u578b\u548c\u63a8\u7406\u6e90\u4ee3\u7801\u5df2\u516c\u5f00\u53d1\u5e03\u3002\u8fd9\u4e00\u4e3e\u63aa\u4e0d\u4ec5\u4e3a\u7814\u7a76\u4eba\u5458\u63d0\u4f9b\u4e86\u5b9d\u8d35\u7684\u7814\u7a76\u57fa\u7840\uff0c\u4e5f\u4e3a\u5f00\u6e90\u793e\u533a\u6ce8\u5165\u4e86\u65b0\u7684\u6d3b\u529b\uff0c\u9f13\u52b1\u66f4\u591a\u521b\u65b0\u4e0e\u5408\u4f5c\u3002\u901a\u8fc7\u5171\u4eab\u6280\u672f\u6210\u679c\uff0c\u6211\u4eec\u5e0c\u671b\u80fd\u591f\u5438\u5f15\u66f4\u591a\u7684\u5f00\u53d1\u8005\u548c\u7814\u7a76\u673a\u6784\u53c2\u4e0e\u5230\u4e1c\u65b9\u8bed\u8a00\u8bed\u97f3\u8bc6\u522b\u7684\u7814\u7a76\u4e2d\u6765\uff0c\u5171\u540c\u63a8\u52a8\u6280\u672f\u7684\u8fdb\u6b65\u3002&nbsp;<\/p>\n\n\n\n<p>&nbsp;Dolphin\uff0c\u4e00\u4e2a\u5927\u89c4\u6a21\u591a\u8bed\u8a00\u591a\u4efb\u52a1\u81ea\u52a8\u8bed\u97f3\u8bc6\u522b (ASR) \u6a21\u578b\u3002Dolphin \u6784\u5efa\u4e8e Whisper \u98ce\u683c\u7684\u67b6\u6784\u4e4b\u4e0a\uff0c\u5e76\u57fa\u4e8e OWSM\uff0c\u96c6\u6210\u4e86\u4e13\u6709\u548c\u516c\u5f00\u53ef\u7528\u7684\u6570\u636e\u96c6\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cDolphin \u5728\u5404\u79cd\u8bed\u8a00\u548c\u6a21\u578b\u89c4\u6a21\u4e0a\u59cb\u7ec8\u4f18\u4e8e\u73b0\u6709\u7684 SOTA \u6a21\u578b\uff0c\u6709\u6548\u5f25\u5408\u4e86\u4e1c\u897f\u65b9\u8bed\u8a00\u4e4b\u95f4\u7684\u6027\u80fd\u5dee\u8ddd\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0cDolphin \u57fa\u7840\u6a21\u578b\u7684\u6027\u80fd\u751a\u81f3\u4f18\u4e8e Whisper large-v3 \u7248\u672c\u3002\u901a\u8fc7\u5f00\u6e90 Dolphin \u57fa\u7840\u6a21\u578b\u3001\u5c0f\u578b\u6a21\u578b\u4ee5\u53ca\u63a8\u7406\u4ee3\u7801\uff0c\u6211\u4eec\u65e8\u5728\u4e3a\u591a\u8bed\u8a00\u8bed\u97f3\u5904\u7406\u7684\u8fdb\u4e00\u6b65\u53d1\u5c55\u505a\u51fa\u8d21\u732e\u3002<\/p>\n\n\n\n<h2>\u652f\u6301\u7684\u8bed\u8a00\u5217\u8868\uff1a<\/h2>\n\n\n\n<h3> Language  code<\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table><thead><tr><th>Language Code<\/th><th>English Name<\/th><th>Chinese Name<\/th><\/tr><\/thead><tbody><tr><td>zh<\/td><td>Mandarin Chinese<\/td><td>\u4e2d\u6587<\/td><\/tr><tr><td>ja<\/td><td>Japanese<\/td><td>\u65e5\u8bed<\/td><\/tr><tr><td>th<\/td><td>Thai<\/td><td>\u6cf0\u8bed<\/td><\/tr><tr><td>ru<\/td><td>Russian<\/td><td>\u4fc4\u8bed<\/td><\/tr><tr><td>ko<\/td><td>Korean<\/td><td>\u97e9\u8bed<\/td><\/tr><tr><td>id<\/td><td>Indonesian<\/td><td>\u5370\u5ea6\u5c3c\u897f\u4e9a\u8bed<\/td><\/tr><tr><td>vi<\/td><td>Vietnamese<\/td><td>\u8d8a\u5357\u8bed<\/td><\/tr><tr><td>ct<\/td><td>Yue Chinese<\/td><td>\u7ca4\u8bed<\/td><\/tr><tr><td>hi<\/td><td>Hindi<\/td><td>\u5370\u5730\u8bed<\/td><\/tr><tr><td>ur<\/td><td>Urdu<\/td><td>\u4e4c\u5c14\u90fd\u8bed<\/td><\/tr><tr><td>ms<\/td><td>Malay<\/td><td>\u9a6c\u6765\u8bed<\/td><\/tr><tr><td>uz<\/td><td>Uzbek<\/td><td>\u4e4c\u5179\u522b\u514b\u8bed<\/td><\/tr><tr><td>ar<\/td><td>Arabic<\/td><td>\u963f\u62c9\u4f2f\u8bed<\/td><\/tr><tr><td>fa<\/td><td>Persian<\/td><td>\u6ce2\u65af\u8bed<\/td><\/tr><tr><td>bn<\/td><td>Bengali<\/td><td>\u5b5f\u52a0\u62c9\u8bed<\/td><\/tr><tr><td>ta<\/td><td>Tamil<\/td><td>\u6cf0\u7c73\u5c14\u8bed<\/td><\/tr><tr><td>te<\/td><td>Telugu<\/td><td>\u6cf0\u5362\u56fa\u8bed<\/td><\/tr><tr><td>ug<\/td><td>Uighur<\/td><td>\u7ef4\u543e\u5c14\u8bed<\/td><\/tr><tr><td>gu<\/td><td>Gujarati<\/td><td>\u53e4\u5409\u62c9\u7279\u8bed<\/td><\/tr><tr><td>my<\/td><td>Burmese<\/td><td>\u7f05\u7538\u8bed<\/td><\/tr><tr><td>tl<\/td><td>Tagalog<\/td><td>\u5854\u52a0\u6d1b\u8bed<\/td><\/tr><tr><td>kk<\/td><td>Kazakh<\/td><td>\u54c8\u8428\u514b\u8bed<\/td><\/tr><tr><td>or<\/td><td>Oriya \/ Odia<\/td><td>\u5965\u91cc\u4e9a\u8bed<\/td><\/tr><tr><td>ne<\/td><td>Nepali<\/td><td>\u5c3c\u6cca\u5c14\u8bed<\/td><\/tr><tr><td>mn<\/td><td>Mongolian<\/td><td>\u8499\u53e4\u8bed<\/td><\/tr><tr><td>km<\/td><td>Khmer<\/td><td>\u9ad8\u68c9\u8bed<\/td><\/tr><tr><td>jv<\/td><td>Javanese<\/td><td>\u722a\u54c7\u8bed<\/td><\/tr><tr><td>lo<\/td><td>Lao<\/td><td>\u8001\u631d\u8bed<\/td><\/tr><tr><td>si<\/td><td>Sinhala<\/td><td>\u50e7\u4f3d\u7f57\u8bed<\/td><\/tr><tr><td>fil<\/td><td>Filipino<\/td><td>\u83f2\u5f8b\u5bbe\u8bed<\/td><\/tr><tr><td>ps<\/td><td>Pushto<\/td><td>\u666e\u4ec0\u56fe\u8bed<\/td><\/tr><tr><td>pa<\/td><td>Panjabi<\/td><td>\u65c1\u906e\u666e\u8bed<\/td><\/tr><tr><td>kab<\/td><td>Kabyle<\/td><td>\u5361\u62dc\u5c14\u8bed<\/td><\/tr><tr><td>ba<\/td><td>Bashkir<\/td><td>\u5df4\u4ec0\u57fa\u5c14\u8bed<\/td><\/tr><tr><td>ks<\/td><td>Kashmiri<\/td><td>\u514b\u4ec0\u7c73\u5c14\u8bed<\/td><\/tr><tr><td>tg<\/td><td>Tajik<\/td><td>\u5854\u5409\u514b\u8bed<\/td><\/tr><tr><td>su<\/td><td>Sundanese<\/td><td>\u5dfd\u4ed6\u8bed<\/td><\/tr><tr><td>mr<\/td><td>Marathi<\/td><td>\u9a6c\u62c9\u5730\u8bed<\/td><\/tr><tr><td>ky<\/td><td>Kirghiz<\/td><td>\u5409\u5c14\u5409\u65af\u8bed<\/td><\/tr><tr><td>az<\/td><td>Azerbaijani<\/td><td>\u963f\u585e\u62dc\u7586\u8bed<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h3 id=\"user-content-language-region-code\">Language Region Code<a href=\"https:\/\/openi.pcl.ac.cn\/DataoceanAI\/Dolphin\/src\/branch\/main\/languages.md#user-content-language-region-code\"><\/a><\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table><thead><tr><th>Language Region Code<\/th><th>English Name<\/th><th>Chinese Name<\/th><\/tr><\/thead><tbody><tr><td>zh-CN<\/td><td>Chinese (Mandarin)<\/td><td>\u4e2d\u6587(\u666e\u901a\u8bdd)<\/td><\/tr><tr><td>zh-TW<\/td><td>Chinese (Taiwan)<\/td><td>\u4e2d\u6587(\u53f0\u6e7e)<\/td><\/tr><tr><td>zh-WU<\/td><td>Chinese (Wuyu)<\/td><td>\u4e2d\u6587(\u5434\u8bed)<\/td><\/tr><tr><td>zh-SICHUAN<\/td><td>Chinese (Sichuan)<\/td><td>\u4e2d\u6587(\u56db\u5ddd\u8bdd)<\/td><\/tr><tr><td>zh-SHANXI<\/td><td>Chinese (Shanxi)<\/td><td>\u4e2d\u6587(\u5c71\u897f\u8bdd)<\/td><\/tr><tr><td>zh-ANHUI<\/td><td>Chinese (Anhui)<\/td><td>\u4e2d\u6587(\u5b89\u5fbd\u8bdd)<\/td><\/tr><tr><td>zh-TIANJIN<\/td><td>Chinese (Tianjin)<\/td><td>\u4e2d\u6587(\u5929\u6d25\u8bdd)<\/td><\/tr><tr><td>zh-NINGXIA<\/td><td>Chinese (Ningxia)<\/td><td>\u4e2d\u6587(\u5b81\u590f\u8bdd)<\/td><\/tr><tr><td>zh-SHAANXI<\/td><td>Chinese (Shaanxi)<\/td><td>\u4e2d\u6587(\u9655\u897f\u8bdd)<\/td><\/tr><tr><td>zh-HEBEI<\/td><td>Chinese (Hebei)<\/td><td>\u4e2d\u6587(\u6cb3\u5317\u8bdd)<\/td><\/tr><tr><td>zh-SHANDONG<\/td><td>Chinese (Shandong)<\/td><td>\u4e2d\u6587(\u5c71\u4e1c\u8bdd)<\/td><\/tr><tr><td>zh-GUANGDONG<\/td><td>Chinese (Guangdong)<\/td><td>\u4e2d\u6587(\u5e7f\u4e1c\u8bdd)<\/td><\/tr><tr><td>zh-SHANGHAI<\/td><td>Chinese (Shanghai)<\/td><td>\u4e2d\u6587(\u4e0a\u6d77\u8bdd)<\/td><\/tr><tr><td>zh-HUBEI<\/td><td>Chinese (Hubei)<\/td><td>\u4e2d\u6587(\u6e56\u5317\u8bdd)<\/td><\/tr><tr><td>zh-LIAONING<\/td><td>Chinese (Liaoning)<\/td><td>\u4e2d\u6587(\u8fbd\u5b81\u8bdd)<\/td><\/tr><tr><td>zh-GANSU<\/td><td>Chinese (Gansu)<\/td><td>\u4e2d\u6587(\u7518\u8083\u8bdd)<\/td><\/tr><tr><td>zh-FUJIAN<\/td><td>Chinese (Fujian)<\/td><td>\u4e2d\u6587(\u798f\u5efa\u8bdd)<\/td><\/tr><tr><td>zh-HUNAN<\/td><td>Chinese (Hunan)<\/td><td>\u4e2d\u6587(\u6e56\u5357\u8bdd)<\/td><\/tr><tr><td>zh-HENAN<\/td><td>Chinese (Henan)<\/td><td>\u4e2d\u6587(\u6cb3\u5357\u8bdd)<\/td><\/tr><tr><td>zh-YUNNAN<\/td><td>Chinese (Yunnan)<\/td><td>\u4e2d\u6587(\u4e91\u5357\u8bdd)<\/td><\/tr><tr><td>zh-MINNAN<\/td><td>Chinese (Minnan)<\/td><td>\u4e2d\u6587(\u95fd\u5357\u8bed)<\/td><\/tr><tr><td>zh-WENZHOU<\/td><td>Chinese (Wenzhou)<\/td><td>\u4e2d\u6587(\u6e29\u5dde\u8bdd)<\/td><\/tr><tr><td>ja-JP<\/td><td>Japanese<\/td><td>\u65e5\u8bed<\/td><\/tr><tr><td>th-TH<\/td><td>Thai<\/td><td>\u6cf0\u8bed<\/td><\/tr><tr><td>ru-RU<\/td><td>Russian<\/td><td>\u4fc4\u8bed<\/td><\/tr><tr><td>ko-KR<\/td><td>Korean<\/td><td>\u97e9\u8bed<\/td><\/tr><tr><td>id-ID<\/td><td>Indonesian<\/td><td>\u5370\u5ea6\u5c3c\u897f\u4e9a\u8bed<\/td><\/tr><tr><td>vi-VN<\/td><td>Vietnamese<\/td><td>\u8d8a\u5357\u8bed<\/td><\/tr><tr><td>ct-NULL<\/td><td>Yue (Unknown)<\/td><td>\u7ca4\u8bed(\u672a\u77e5)<\/td><\/tr><tr><td>ct-HK<\/td><td>Yue (Hongkong)<\/td><td>\u7ca4\u8bed(\u9999\u6e2f)<\/td><\/tr><tr><td>ct-GZ<\/td><td>Yue (Guangdong)<\/td><td>\u7ca4\u8bed(\u5e7f\u4e1c)<\/td><\/tr><tr><td>hi-IN<\/td><td>Hindi<\/td><td>\u5370\u5730\u8bed<\/td><\/tr><tr><td>ur-IN<\/td><td>Urdu<\/td><td>\u4e4c\u5c14\u90fd\u8bed(\u5370\u5ea6)<\/td><\/tr><tr><td>ur-PK<\/td><td>Urdu (Islamic Republic of Pakistan)<\/td><td>\u4e4c\u5c14\u90fd\u8bed<\/td><\/tr><tr><td>ms-MY<\/td><td>Malay<\/td><td>\u9a6c\u6765\u8bed<\/td><\/tr><tr><td>uz-UZ<\/td><td>Uzbek<\/td><td>\u4e4c\u5179\u522b\u514b\u8bed<\/td><\/tr><tr><td>ar-MA<\/td><td>Arabic (Morocco)<\/td><td>\u963f\u62c9\u4f2f\u8bed(\u6469\u6d1b\u54e5)<\/td><\/tr><tr><td>ar-GLA<\/td><td>Arabic<\/td><td>\u963f\u62c9\u4f2f\u8bed<\/td><\/tr><tr><td>ar-SA<\/td><td>Arabic (Saudi Arabia)<\/td><td>\u963f\u62c9\u4f2f\u8bed(\u6c99\u7279)<\/td><\/tr><tr><td>ar-EG<\/td><td>Arabic (Egypt)<\/td><td>\u963f\u62c9\u4f2f\u8bed(\u57c3\u53ca)<\/td><\/tr><tr><td>ar-KW<\/td><td>Arabic (Kuwait)<\/td><td>\u963f\u62c9\u4f2f\u8bed(\u79d1\u5a01\u7279)<\/td><\/tr><tr><td>ar-LY<\/td><td>Arabic (Libya)<\/td><td>\u963f\u62c9\u4f2f\u8bed(\u5229\u6bd4\u4e9a)<\/td><\/tr><tr><td>ar-JO<\/td><td>Arabic (Jordan)<\/td><td>\u963f\u62c9\u4f2f\u8bed(\u7ea6\u65e6)<\/td><\/tr><tr><td>ar-AE<\/td><td>Arabic (U.A.E.)<\/td><td>\u963f\u62c9\u4f2f\u8bed(\u963f\u8054\u914b)<\/td><\/tr><tr><td>ar-LVT<\/td><td>Arabic (Levant)<\/td><td>\u963f\u62c9\u4f2f\u8bed(\u9ece\u51e1\u7279)<\/td><\/tr><tr><td>fa-IR<\/td><td>Persian<\/td><td>\u6ce2\u65af\u8bed<\/td><\/tr><tr><td>bn-BD<\/td><td>Bengali<\/td><td>\u5b5f\u52a0\u62c9\u8bed<\/td><\/tr><tr><td>ta-SG<\/td><td>Tamil (Singaporean)<\/td><td>\u6cf0\u7c73\u5c14\u8bed(\u65b0\u52a0\u5761)<\/td><\/tr><tr><td>ta-LK<\/td><td>Tamil (Sri Lankan)<\/td><td>\u6cf0\u7c73\u5c14\u8bed(\u65af\u91cc\u5170\u5361)<\/td><\/tr><tr><td>ta-IN<\/td><td>Tamil (India)<\/td><td>\u6cf0\u7c73\u5c14\u8bed(\u5370\u5ea6)<\/td><\/tr><tr><td>ta-MY<\/td><td>Tamil (Malaysia)<\/td><td>\u6cf0\u7c73\u5c14\u8bed(\u9a6c\u6765\u897f\u4e9a)<\/td><\/tr><tr><td>te-IN<\/td><td>Telugu<\/td><td>\u6cf0\u5362\u56fa\u8bed<\/td><\/tr><tr><td>ug-NULL<\/td><td>Uighur<\/td><td>\u7ef4\u543e\u5c14\u8bed<\/td><\/tr><tr><td>ug-CN<\/td><td>Uighur<\/td><td>\u7ef4\u543e\u5c14\u8bed<\/td><\/tr><tr><td>gu-IN<\/td><td>Gujarati<\/td><td>\u53e4\u5409\u62c9\u7279\u8bed<\/td><\/tr><tr><td>my-MM<\/td><td>Burmese<\/td><td>\u7f05\u7538\u8bed<\/td><\/tr><tr><td>tl-PH<\/td><td>Tagalog<\/td><td>\u5854\u52a0\u6d1b\u8bed<\/td><\/tr><tr><td>kk-KZ<\/td><td>Kazakh<\/td><td>\u54c8\u8428\u514b\u8bed<\/td><\/tr><tr><td>or-IN<\/td><td>Oriya \/ Odia<\/td><td>\u5965\u91cc\u4e9a\u8bed<\/td><\/tr><tr><td>ne-NP<\/td><td>Nepali<\/td><td>\u5c3c\u6cca\u5c14\u8bed<\/td><\/tr><tr><td>mn-MN<\/td><td>Mongolian<\/td><td>\u8499\u53e4\u8bed<\/td><\/tr><tr><td>km-KH<\/td><td>Khmer<\/td><td>\u9ad8\u68c9\u8bed<\/td><\/tr><tr><td>jv-ID<\/td><td>Javanese<\/td><td>\u722a\u54c7\u8bed<\/td><\/tr><tr><td>lo-LA<\/td><td>Lao<\/td><td>\u8001\u631d\u8bed<\/td><\/tr><tr><td>si-LK<\/td><td>Sinhala<\/td><td>\u50e7\u4f3d\u7f57\u8bed<\/td><\/tr><tr><td>fil-PH<\/td><td>Filipino<\/td><td>\u83f2\u5f8b\u5bbe\u8bed<\/td><\/tr><tr><td>ps-AF<\/td><td>Pushto<\/td><td>\u666e\u4ec0\u56fe\u8bed<\/td><\/tr><tr><td>pa-IN<\/td><td>Panjabi<\/td><td>\u65c1\u906e\u666e\u8bed<\/td><\/tr><tr><td>kab-NULL<\/td><td>Kabyle<\/td><td>\u5361\u62dc\u5c14\u8bed<\/td><\/tr><tr><td>ba-NULL<\/td><td>Bashkir<\/td><td>\u5df4\u4ec0\u57fa\u5c14\u8bed<\/td><\/tr><tr><td>ks-IN<\/td><td>Kashmiri<\/td><td>\u514b\u4ec0\u7c73\u5c14\u8bed<\/td><\/tr><tr><td>tg-TJ<\/td><td>Tajik<\/td><td>\u5854\u5409\u514b\u8bed<\/td><\/tr><tr><td>su-ID<\/td><td>Sundanese<\/td><td>\u5dfd\u4ed6\u8bed<\/td><\/tr><tr><td>mr-IN<\/td><td>Marathi<\/td><td>\u9a6c\u62c9\u5730\u8bed<\/td><\/tr><tr><td>ky-KG<\/td><td>Kirghiz<\/td><td>\u5409\u5c14\u5409\u65af\u8bed<\/td><\/tr><tr><td>az-AZ<\/td><td>Azerbaijani<\/td><td>\u963f\u585e\u62dc\u7586\u8bed<\/td><\/tr><\/tbody><\/table><\/figure>\n","protected":false},"excerpt":{"rendered":"<p>\u8bba\u6587\u9898\u76ee\uff1aDolphin: A Large-Scale Automatic Speech Recognitio &hellip; <a href=\"http:\/\/139.9.1.231\/index.php\/2025\/04\/22\/dolphin-asr-model-llm\/\" class=\"more-link\">\u7ee7\u7eed\u9605\u8bfb<span class=\"screen-reader-text\">Dolphin -\u652f\u6301\u4e1c\u65b940\u8bed\u79cd+\u4e2d\u56fd22\u65b9\u8a00\u7684\u65b0SOTA\u8bed\u97f3\u5927\u6a21\u578b<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[21,4,9,38,34],"tags":[],"_links":{"self":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/25807"}],"collection":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/comments?post=25807"}],"version-history":[{"count":65,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/25807\/revisions"}],"predecessor-version":[{"id":26338,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/25807\/revisions\/26338"}],"wp:attachment":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/media?parent=25807"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/categories?post=25807"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/tags?post=25807"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}