{"id":53760,"date":"2025-02-16T12:43:11","date_gmt":"2025-02-16T04:43:11","guid":{"rendered":"https:\/\/fwq.ai\/blog\/53760\/"},"modified":"2025-02-16T12:43:11","modified_gmt":"2025-02-16T04:43:11","slug":"%e5%88%b6%e4%bd%9c%e9%ab%98%e8%b4%a8%e9%87%8f%e7%9a%84%e8%af%ad%e9%9f%b3%e5%85%8b%e9%9a%86%e6%95%b0%e6%8d%ae%e9%9b%86","status":"publish","type":"post","link":"https:\/\/fwq.ai\/blog\/53760\/","title":{"rendered":"\u5236\u4f5c\u9ad8\u8d28\u91cf\u7684\u8bed\u97f3\u514b\u9686\u6570\u636e\u96c6"},"content":{"rendered":"<blockquote><p>\n  \u91cd\u8981\u63d0\u793a\uff1a\u672c\u6307\u5357\u4ec5\u4f9b\u7b26\u5408\u9053\u5fb7\u89c4\u8303\u7684\u4f7f\u7528\u3002\u8bf7\u786e\u4fdd\u4f60\u62e5\u6709\u7528\u4e8e\u6570\u636e\u96c6\u521b\u5efa\u7684\u4efb\u4f55\u5a92\u4f53\u7684\u5fc5\u8981\u6743\u9650\uff0c\u5e76\u9075\u5b88\u9002\u7528\u7684\u6cd5\u5f8b\u548c\u51c6\u5219\u3002\u4f7f\u7528 YouTube-DL \u7b49\u5de5\u5177\u65f6\uff0c\u6211\u4eec\u5efa\u8bae\u4e0b\u8f7d\u548c\u5904\u7406\u4f60\u62e5\u6709\u6240\u6709\u6743\u7684\u5185\u5bb9\uff08\u4f8b\u5982\u60a8\u81ea\u5df1\u7684\u5f55\u97f3\uff09\uff0c\u4ee5\u5c0a\u91cd\u4ed6\u4eba\u7684\u77e5\u8bc6\u4ea7\u6743\u3002\u8bed\u97f3\u514b\u9686\u6280\u672f\u5e94\u59cb\u7ec8\u4ee5\u8d1f\u8d23\u4efb\u7684\u65b9\u5f0f\u4f7f\u7528\uff0c\u5e76\u627f\u8bfa\u4fdd\u62a4\u9690\u79c1\u5e76\u5c0a\u91cd\u4e2a\u4eba\u3002\n<\/p><\/blockquote>\n<p>\u968f\u7740\u63d0\u4f9b\u914d\u97f3\u548c\u7c7b\u4f3c\u4eba\u7c7b\u5bf9\u8bdd\u58f0\u97f3\u7684 AI \u5e94\u7528\u7a0b\u5e8f\u7684\u5174\u8d77\uff0c\u4eba\u4eec\u5bf9\u6784\u5efa\u81ea\u5b9a\u4e49\u6587\u672c\u8f6c\u8bed\u97f3\u6a21\u578b\u7684\u5174\u8da3\u65e5\u76ca\u6d53\u539a\u3002\u8bb8\u591a\u5f00\u53d1\u4eba\u5458\u548c\u516c\u53f8\u8bd5\u56fe\u901a\u8fc7\u5fae\u8c03\u81ea\u5df1\u7684\u6a21\u578b\u6765\u907f\u514d\u4ed8\u8d39\u914d\u97f3\u670d\u52a1\u7684\u6210\u672c\u3002\u7136\u800c\uff0c\u521b\u5efa\u9ad8\u8d28\u91cf\u6587\u672c\u8f6c\u8bed\u97f3\u7cfb\u7edf\u7684\u7b2c\u4e00\u6b65\u4e5f\u662f\u6700\u5173\u952e\u7684\u4e00\u6b65\u662f\u83b7\u53d6\u4e30\u5bcc\u3001\u51c6\u5907\u5145\u5206\u7684\u6570\u636e\u96c6\u3002\u672c\u6307\u5357\u4ecb\u7ecd\u4e86\u6784\u5efa\u6b64\u7c7b\u6570\u636e\u96c6\u7684\u5168\u9762\u8fc7\u7a0b\uff0c\u91cd\u70b9\u662f\u63d0\u53d6\u6709\u6548\u8bed\u97f3\u514b\u9686\u6240\u5fc5\u9700\u7684\u6e05\u6670\u51c6\u786e\u7684\u58f0\u97f3\u6837\u672c\u3002<\/p>\n<p>\u4f5c\u4e3a\u6211\u9879\u76ee\u7684\u4e2d\u95f4\u6b65\u9aa4\uff0c\u6211\u6b63\u5728\u52aa\u529b\u5efa\u7acb\u4e00\u4e2a\u53ef\u4ee5\u65e0\u7f1d\u6267\u884c\u6bcf\u4e2a\u6b65\u9aa4\u7684\u81ea\u52a8\u5316\u7ba1\u9053\u3002\u672c\u6307\u5357\u63d0\u4f9b\u4e86\u6709\u5173\u521b\u5efa\u9ad8\u8d28\u91cf\u6570\u636e\u96c6\u7684\u8be6\u7ec6\u6f14\u7ec3\uff0c\u6db5\u76d6\u4e86\u4ece\u89c6\u9891\u4e0b\u8f7d\u5230\u97f3\u9891\u8f6c\u5f55\u7684\u6240\u6709\u5185\u5bb9\u3002\u6700\u540e\uff0c\u6211\u5c06\u8ba8\u8bba\u5728\u6b64\u8fc7\u7a0b\u4e2d\u9047\u5230\u7684\u4e00\u4e9b\u6311\u6218\u3002<\/p>\n<p>  \u8bed\u97f3\u63d0\u53d6\u6982\u8ff0 <\/p>\n<p>\u6b64\u5de5\u4f5c\u6d41\u7a0b\u4f7f\u7528 YouTube-DL\u3001FFMPEG\u3001WhisperX \u7b49\u5de5\u5177\u6765\u786e\u4fdd\u63d0\u53d6\u7684\u97f3\u9891\u7b26\u5408\u9ad8\u8d28\u91cf\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684\u8981\u6c42\u3002\u8ba9\u6211\u4eec\u6df1\u5165\u4e86\u89e3\u6bcf\u4e2a\u6b65\u9aa4\u4ee5\u53ca\u5e2e\u52a9\u60a8\u5165\u95e8\u7684\u5fc5\u8981\u547d\u4ee4\u3002<\/p>\n<h2>1\u3001\u4e0b\u8f7d\u97f3\u9891\u6e90<\/h2>\n<p>\u7b2c\u4e00\u6b65\u662f\u4ece CSV \u6587\u4ef6\u4e2d\u63d0\u4f9b\u7684 URL \u5217\u8868\u4e2d\u4e0b\u8f7d\u89c6\u9891\u3002 \u662f\u4e00\u4e2a\u7b80\u5316\u6b64\u8fc7\u7a0b\u7684\u547d\u4ee4\u884c\u5de5\u5177\u3002\u5b83\u5141\u8bb8\u60a8\u4e0b\u8f7d\u9ad8\u8d28\u91cf\u7684\u89c6\u9891\u5e76\u652f\u6301\u5404\u79cd\u89c6\u9891\u6258\u7ba1\u5e73\u53f0\u3002<\/p>\n<ul>\n<li>\u4ece CSV \u52a0\u8f7d URL\uff1a\u4ece CSV \u6587\u4ef6\u4e2d\u63d0\u53d6 URL \u5217\u8868\u3002<\/li>\n<li>\u8fd0\u884c YouTube-DL\uff1a\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u4ee5\u6700\u4f73\u53ef\u7528\u8d28\u91cf\u4e0b\u8f7d\u89c6\u9891\u3002<\/li>\n<\/ul>\n<p>\u7528\u6cd5\uff1a<\/p>\n<pre><code>youtube-dl -f 'bestvideo[ext=mp4]+bestaudio[ext=m4a]\/mp4' &lt;URL&gt;<\/code><\/pre>\n<blockquote><p>\n  \u6ce8\u610f\uff1a\u4ece 2024 \u5e74 8 \u6708\u5f00\u59cb\uff0cyoutube-dl \u8981\u6c42\u4f60\u5728\u4ece AWS EC2 \u8fd0\u884c\u65f6\u4f7f\u7528 OAuth\/Cookies \u4f20\u9012\u8eab\u4efd\u9a8c\u8bc1\u8be6\u7ec6\u4fe1\u606f\uff0c\u56e0\u4e3a Google \u5bf9\u4e0b\u8f7d T&amp;C \u975e\u5e38\u4e25\u683c\uff0c\u5e76\u5c06\u8bb8\u591a\u516c\u5171\u4e91\u63d0\u4f9b\u5546 IP \u5217\u5165\u9ed1\u540d\u5355\u3002\n<\/p><\/blockquote>\n<h2>2\u3001\u5206\u79bb\u97f3\u89c6\u9891<\/h2>\n<p>\u4e0b\u8f7d\u89c6\u9891\u540e\uff0c\u4e0b\u4e00\u6b65\u662f\u5c06\u97f3\u8f68\u4e0e\u89c6\u9891\u6587\u4ef6\u5206\u79bb\u3002 \u662f\u4e00\u4e2a\u529f\u80fd\u5f3a\u5927\u7684\u5de5\u5177\uff0c\u53ef\u4ee5\u8f7b\u677e\u63d0\u53d6\u548c\u8f6c\u6362\u97f3\u9891\u683c\u5f0f\u3002\u5c06\u97f3\u9891\u8f6c\u6362\u4e3a WAV \u683c\u5f0f\u53ef\u786e\u4fdd\u4e0e\u5927\u591a\u6570\u673a\u5668\u5b66\u4e60\u6846\u67b6\u517c\u5bb9\u3002<\/p>\n<p>\u7528\u6cd5\uff1a<\/p>\n<pre><code>ffmpeg -i input_video.mp4 -q:a 0 -map a output_audio.wav<\/code><\/pre>\n<p>\u5728\u8fd9\u91cc\uff0c <code>-map a<\/code> \u9694\u79bb\u97f3\u8f68\uff0c <code>-q:a 0<\/code> \u4fdd\u7559\u97f3\u9891\u8d28\u91cf\u3002<\/p>\n<h2>3\u3001\u63d0\u53d6\u4eba\u58f0<\/h2>\n<p>\u9694\u79bb\u97f3\u9891\u540e\uff0c\u4e0b\u4e00\u6b65\u662f\u4e13\u6ce8\u4e8e\u4ece\u97f3\u9891\u4e2d\u4ec5\u63d0\u53d6\u4eba\u58f0\u6210\u5206\u3002\u80cc\u666f\u97f3\u4e50\u548c\u566a\u97f3\u4f1a\u5e72\u6270\u8bed\u97f3\u514b\u9686\u3002\u5728\u4ece\u97f3\u8f68\u4e2d\u9694\u79bb\u4eba\u58f0\u65b9\u9762\uff0c \u662f\u6700\u4f73\u9009\u62e9\u4e4b\u4e00\u3002\u5b83\u5229\u7528\u5148\u8fdb\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\uff0c\u5305\u62ec MDX-Net \u548c Demucs\uff0c\u4ee5\u6700\u5c11\u7684\u4f2a\u5f71\u5b9e\u73b0\u9ad8\u8d28\u91cf\u7684\u4eba\u58f0\u5206\u79bb\u3002<\/p>\n<p>Demucs \u662f UVR \u4e2d\u5f3a\u5927\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\uff0c\u4f7f\u7528\u5377\u79ef\u795e\u7ecf\u7f51\u7edc (CNN) \u5b9e\u73b0\u8de8\u5404\u79cd\u97f3\u4e50\u6d41\u6d3e\u7684\u51c6\u786e\u548c\u7075\u6d3b\u7684\u4eba\u58f0\u5206\u79bb\u3002MDX-Net \u662f\u4e00\u4e2a\u53ef\u9009\u6a21\u578b\uff0c\u4e13\u95e8\u7528\u4e8e\u66f4\u7cbe\u7ec6\u7684\u4eba\u58f0\u63d0\u53d6\uff0c\u5728 Demucs \u53ef\u80fd\u7559\u4e0b\u6b8b\u7559\u4f2a\u5f71\u7684\u60c5\u51b5\u4e0b\u63d0\u4f9b\u66f4\u597d\u7684\u7ed3\u679c\u3002UVR \u8fd8\u652f\u6301\u5176\u4ed6\u6a21\u578b\uff0c\u4f8b\u5982 Ensemble \u6a21\u5f0f\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528\u8be5\u6a21\u5f0f\u7ec4\u5408\u591a\u4e2a\u6a21\u578b\u4ee5\u83b7\u5f97\u66f4\u597d\u7684\u8d28\u91cf\u7ed3\u679c\u3002<\/p>\n<p>\u4f7f\u7528\uff1a<\/p>\n<p>UVR \u7684\u4f7f\u7528\u975e\u5e38\u7b80\u5355\uff0c\u5b83\u9996\u5148\u8981\u6c42\u4f60\u9009\u62e9\u4ee5\u4e0b\u4efb\u4f55\u4e00\u79cd\u5904\u7406\u65b9\u6cd5\uff1a<\/p>\n<p>\u7136\u540e\u5728\u65b9\u6cd5\u4e2d\u9009\u62e9\u4efb\u4f55\u7279\u5b9a\u7684\u53d7\u652f\u6301\u6a21\u578b\u3002\u4f46\u662f\u5728\u4f7f\u7528\u4e4b\u524d\uff0c\u4f60\u5fc5\u987b\u4ece UI \u4e0b\u8f7d\u8fd9\u4e9b\u6a21\u578b\u3002\u4ee5\u4e0b\u662f\u6211\u6309\u7167\u6709\u6548\u8bed\u97f3\u63d0\u53d6\u7684\u987a\u5e8f\u63a8\u8350\u7684\u65b9\u6cd5+\u6a21\u578b\u7ec4\u5408\uff1a<\/p>\n<pre><code>Ensemble -&gt; Demucs-&gt; MDX-Net + Kim vocals 2<\/code><\/pre>\n<p>  \u8bed\u97f3\u63d0\u53d6\u7684\u4e94\u6b65\u6d41\u7a0b <\/p>\n<h2>4\u3001\u8bf4\u8bdd\u4eba\u5206\u7c7b<\/h2>\n<p>\u4e3a\u4e86\u521b\u5efa\u5177\u6709\u4e0d\u540c\u8bf4\u8bdd\u4eba\u6807\u7b7e\u7684\u6570\u636e\u96c6\uff0c\u6211\u4eec\u4f7f\u7528\u8bf4\u8bdd\u4eba\u5206\u7c7b\u3002\u6b64\u8fc7\u7a0b\u6309\u8bf4\u8bdd\u4eba\u8bc6\u522b\u548c\u5206\u5272\u97f3\u9891\u90e8\u5206\uff0c\u8fd9\u5728\u5904\u7406\u591a\u8bf4\u8bdd\u4eba\u6570\u636e\u65f6\u7279\u522b\u6709\u7528\u3002 \u5728 Hugging face \u4e0a\u6709\u9ad8\u6548\u7684\u8bf4\u8bdd\u4eba\u5206\u7c7b\u6a21\u578b\uff0c\u53ef\u4ee5\u968f\u65f6\u7528\u4e8e\u63d0\u53d6\u4eba\u58f0\u4e2d\u7684\u591a\u4e2a\u8bf4\u8bdd\u4eba\u3002<\/p>\n<p>\u7528\u6cd5\uff1a<\/p>\n<pre><code>from pyannote.audio import Pipeline\npipeline = Pipeline.from_pretrained(\"pyannote\/speaker-diarization@2.1\",\n                                    use_auth_token=\"ACCESS_TOKEN_GOES_HERE\")\n\n# apply the pipeline to an audio file\ndiarization = pipeline(\"audio.wav\")\n# dump the diarization output to disk using RTTM format\nwith open(\"audio.rttm\", \"w\") as rttm:\n    diarization.write_rttm(rttm)<\/code><\/pre>\n<p>\u8bf7\u786e\u4fdd\u4f60\u751f\u6210 HF \u8bbf\u95ee\u4ee4\u724c\uff0c\u5e76\u901a\u8fc7\u63a5\u53d7\u7684 T&amp;C \u83b7\u5f97\u8bfb\u53d6\u8bbf\u95ee\u6743\u9650\u3002\u5982\u679c\u6ca1\u6709\u8fd9\u4e2a\uff0c\u4f60\u5c06\u6536\u5230\u9519\u8bef \u6b64\u811a\u672c\u5c06\u521b\u5efa\u4e00\u4e2a <code>.rttm<\/code> \u6587\u4ef6\uff0c\u5176\u4e2d\u5305\u542b\u6bcf\u4e2a\u8bf4\u8bdd\u8005\u7684\u65f6\u95f4\u6bb5\uff0c\u7a0d\u540e\u53ef\u7528\u4e8e\u6309\u8bf4\u8bdd\u8005\u6807\u8bb0\u97f3\u9891\u7247\u6bb5\u3002<\/p>\n<p>\u6216\u8005\uff0c\u6211\u4eec\u4e5f\u53ef\u4ee5\u4f7f\u7528 WhisperX \u5305\u88c5\u5668\u4e2d\u7684 \u6a21\u578b\u3002\u6211\u4eec\u8fd8\u53ef\u4ee5\u4f7f\u7528 WhisperX \u51c6\u5907\u97f3\u9891\u6587\u4ef6\u7684\u8f6c\u5f55\uff0c\u5982\u6211\u4eec\u5728\u4e0b\u4e00\u6b65\u4e2d\u770b\u5230\u7684\u90a3\u6837\uff1a<\/p>\n<pre><code>import whisperx\n\ndevice = \"cuda\" # Use cpu if you are running on a GPU less machine\ndef convert_diarize_segments(diarize_segments_dict):\n    result = []\n    for segment in diarize_segments_dict:\n        result.append({\n            'start': segment[\"start\"],\n            'end': segment[\"end\"],\n            'label': segment[\"label\"],\n            'speaker': segment[\"speaker\"]\n        })\n    return result\n\ndef identify_speakers(audio_file:str, min_speakers=1, max_speakers=2):\n    device = \"cuda\" # Use cpu if you do not have GPU\n    audio = whisperx.load_audio(audio_file)\n    diarize_model = whisperx.DiarizationPipeline(\n        use_auth_token=\"ACCESS_TOKEN_GOES_HERE\", device=device)\n    diarize_segments = diarize_model(\n        audio, min_speakers=min_speakers, max_speakers=max_speakers)\n    result_json = convert_diarize_segments(\n        diarize_segments.to_dict(orient='records')\n    return result_json\n\nidentify_speakers(\"&lt;PATH_TO_AUDIO_WAV&gt;\", 1, 2)<\/code><\/pre>\n<h2>5\u3001\u97f3\u9891\u8f6c\u5f55<\/h2>\n<p>\u6700\u540e\uff0c\u8f6c\u5f55\u97f3\u9891\u5bf9\u4e8e\u521b\u5efa\u8bad\u7ec3\u8bed\u97f3\u514b\u9686\u6a21\u578b\u6240\u9700\u7684\u6587\u672c-\u97f3\u9891\u5bf9\u81f3\u5173\u91cd\u8981\u3002WhisperX \u5efa\u7acb\u5728 OpenAI \u7684 Whisper \u6a21\u578b\u4e4b\u4e0a\uff0c\u5373\u4f7f\u5bf9\u4e8e\u5177\u6709\u6311\u6218\u6027\u7684\u97f3\u9891\u8f93\u5165\u4e5f\u80fd\u63d0\u4f9b\u9ad8\u8d28\u91cf\u7684\u8f6c\u5f55\u3002\u60a8\u53ef\u4ee5\u4f7f\u7528 \u6216\u6a21\u578b\u6765\u6267\u884c\u97f3\u9891\u5206\u6790\u3002<\/p>\n<p>\u7528\u6cd5\uff1a<\/p>\n<pre><code>import whisperx\n\ndef get_transcript(audio_file):\n    device = os.environ.get(\"DEVICE\")\n    batch_size = int(os.environ.get(\"BATCH_SIZE\"))\n    compute_type = os.environ.get(\"COMPUTE_TYPE\")\n    model_dir = os.environ.get(\"MODEL_DIR\")\n    model = whisperx.load_model(\n        \"large-v2\", device, compute_type=compute_type, download_root=model_dir)\n    audio = whisperx.load_audio(audio_file)\n    transcript = model.transcribe(audio, batch_size=batch_size)\n    if os.environ.get(\"DELETE_MODEL_AFTER_USE\"):\n        gc.collect()\n        torch.cuda.empty_cache()\n        del model\n    return transcript\n\nget_transcript(\"&lt;PATH_TO_AUDIO_WAV&gt;\")<\/code><\/pre>\n<p>\u6b64\u8f6c\u5f55\u672c\u5c06\u5305\u542b\u6bcf\u4e2a\u8f6c\u5f55\u7684\u8bed\u97f3\u7247\u6bb5\u53ca\u5176\u76f8\u5e94\u7684\u65f6\u95f4\u6233\u3002\u8fd9\u5bf9\u4e8e\u5c06\u6587\u672c\u4e0e\u97f3\u9891\u7247\u6bb5\u914d\u5bf9\u81f3\u5173\u91cd\u8981\u3002<\/p>\n<h2>6\u3001\u6700\u7ec8\u8f93\u51fa\uff1a\u7ec4\u7ec7\u6570\u636e\u96c6<\/h2>\n<p>\u5904\u7406\u5b8c\u6240\u6709\u6587\u4ef6\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u5c06\u5b83\u4eec\u7ec4\u7ec7\u6210\u4e00\u4e2a\u7ed3\u6784\u5316\u7684\u6570\u636e\u96c6\uff0c\u7136\u540e\u53ef\u4ee5\u5c06\u5176\u4f20\u9012\u7ed9\u5fae\u8c03\u4f60\u81ea\u5df1\u7684\u6a21\u578b\uff0c\u4f8b\u5982 \u3002<\/p>\n<pre><code>dataset_folder\/\n\u251c\u2500\u2500 transcript.txt\n\u251c\u2500\u2500 wavs\/\n\u2502   \u251c\u2500\u2500 011c485067be466aa729c5e33bc4fa49_SPEAKER_00.wav\n\u2502   \u251c\u2500\u2500 029b73c7b0174f48911878fb9198864_SPEAKER_00.wav\n\u2502   \u2514\u2500\u2500 0911b943a05d4313abc3350e5f51e1d0_SPEAKER_00.wav<\/code><\/pre>\n<ul>\n<li><code>wavs\/<\/code> \u5305\u542b\u6bcf\u4e2a\u8bf4\u8bdd\u8005\u7247\u6bb5\u7684\u5355\u72ec\u97f3\u9891\u6587\u4ef6\uff0c\u5e76\u6807\u6709\u552f\u4e00\u6807\u8bc6\u7b26\u3002<\/li>\n<li><code>transcript.txt<\/code> \u7528\u4f5c\u6240\u6709\u6570\u636e\u96c6\u6570\u636e\u7684\u5143\u6570\u636e<\/li>\n<\/ul>\n<pre><code>011c485067be466aa729c5e33bc4fa49_SPEAKER_00 |  &lt;transcripted text&gt;\n029b73c7b0174f48911878fb9198864_SPEAKER_00  |   &lt;transcripted text&gt;\n0911b943a05d4313abc3350e5f51e1d0_SPEAKER_00 |   &lt;transcripted text&gt;\n.\n.<\/code><\/pre>\n<h2>7\u3001\u5927\u89c4\u6a21\u8fd0\u884c<\/h2>\n<p>\u867d\u7136\u4e0a\u8ff0\u6b65\u9aa4\u53ef\u7528\u4e8e\u5904\u7406\u5c11\u91cf\u97f3\u9891\uff0c\u4f46\u5982\u679c\u8981\u63d0\u53d6\u5927\u91cf\u97f3\u9891\u6587\u4ef6\uff0c\u5219\u9700\u8981\u6784\u5efa\u4e00\u4e2a\u53ef\u4ee5\u5904\u7406\u5927\u91cf\u6587\u4ef6\u7684\u7ba1\u9053\u3002\u4ee5\u4e0b\u662f\u6211\u5728\u5927\u89c4\u6a21\u6267\u884c\u6b64\u64cd\u4f5c\u65f6\u9047\u5230\u7684\u4e00\u4e9b\u6311\u6218<\/p>\n<ul>\n<li>\u8bbf\u95ee GPU \u2014 \u867d\u7136\u67d0\u4e9b\u6b65\u9aa4 (1-2) \u53ef\u4ee5\u5728\u666e\u901a CPU \u4e0a\u8fd0\u884c\uff0c\u4f46\u8981\u8fd0\u884c\u5176\u4ed6\u6b65\u9aa4\uff0c\u4f60\u9700\u8981\u83b7\u5f97 GPU\u3002\u5982\u679c\u4f60\u6709\u8bbf\u95ee\u6743\u9650\uff0c\u53ef\u4ee5\u5728 RunPOD \u4e0a\u79df\u7528 GPU \u6216\u5728 AWS \u4e0a\u751f\u6210 g4dn.* EC2 \u673a\u5668\u3002\u6700\u4fbf\u5b9c\u7684 EC2 \u673a\u5668\u6bcf\u5c0f\u65f6\u8d39\u7528\u4e0d\u5230 1 \u7f8e\u5143\u3002<\/li>\n<li>\u73af\u5883\u8bbe\u7f6e \u2014 \u8bbf\u95ee GPU \u662f\u4e00\u56de\u4e8b\uff0c\u4f46\u914d\u7f6e\u673a\u5668\u4ee5\u5728\u4e0a\u9762\u8fd0\u884c\u53c8\u662f\u53e6\u4e00\u56de\u4e8b\u3002\u4f60\u9700\u8981\u9996\u5148\u5b89\u88c5\u517c\u5bb9\u7248\u672c\u7684 pytorch \u548c Nvidia \u9a71\u52a8\u7a0b\u5e8f\u3002\u7136\u540e\u83b7\u53d6\u9002\u5f53\u7684\u5e93\u548c\u6a21\u578b\u3002\u5982\u679c\u4f60\u60f3\u5feb\u901f\u8fdb\u884c\u4e00\u4e9b POC\uff0cRunPOD \u673a\u5668\u53ef\u4ee5\u63d0\u4f9b\u5e2e\u52a9\u3002RunPOD \u4e0a\u7684\u4e00\u4e9b\u914d\u7f6e\u968f\u6240\u6709\u8fd9\u4e9b\u8f6f\u4ef6\u548c\u9a71\u52a8\u7a0b\u5e8f\u4e00\u8d77\u5b89\u88c5\u3002\u5173\u4e8e EC2 \u7684\u8bf4\u660e\uff0c\u4f60\u53ef\u80fd\u9700\u8981\u63d0\u51fa\u7279\u6b8a\u914d\u989d\u8bf7\u6c42\u6765\u751f\u6210 g4dn* \u673a\u5668\u3002\u9ed8\u8ba4\u60c5\u51b5\u4e0b\uff0c\u5b83\u4eec\u662f\u7981\u7528\u7684\u3002<\/li>\n<li>\u81ea\u5b9a\u4e49 \u2014 \u4e0a\u8ff0\u4e00\u4e9b\u6b65\u9aa4\u6d89\u53ca\u4f7f\u7528 UI\uff08\u6b65\u9aa4 3\uff09\u6765\u5b8c\u6210\u4efb\u52a1\u3002\u4f46\u662f\u5982\u679c\u4f60\u60f3\u81ea\u52a8\u5316\u5b83\uff0c\u90a3\u4e48\u9700\u8981\u5728\u673a\u5668\u4e0a\u8fd0\u884c\u539f\u59cb\u6a21\u578b\u3002\u4e3a\u6b64\uff0c\u4f60\u9700\u8981\u56f4\u7ed5\u6b64\u7f16\u5199\u4e00\u4e9b\u4ee3\u7801\u6765\u8bbe\u7f6e\u7ba1\u9053\u3002<\/li>\n<\/ul>\n<h2>8\u3001\u7ed3\u675f\u8bed<\/h2>\n<p>\u901a\u8fc7\u9075\u5faa\u8fd9\u4e9b\u6b65\u9aa4\uff0c\u4f60\u53ef\u4ee5\u521b\u5efa\u4e00\u4e2a\u9002\u5408\u8bed\u97f3\u514b\u9686\u7684\u7ed3\u6784\u5316\u548c\u9ad8\u8d28\u91cf\u6570\u636e\u96c6\u3002\u901a\u8fc7\u8fd9\u79cd\u6570\u636e\u96c6\u51c6\u5907\u65b9\u6cd5\uff0c\u4f60\u5c06\u80fd\u591f\u751f\u6210\u5408\u6210\u8bed\u97f3\uff0c<\/p>\n<p>\u6355\u6349\u771f\u5b9e\u4e16\u754c\u8bed\u97f3\u7684\u7ec6\u5fae\u5dee\u522b\uff0c\u89e3\u9501\u4e2a\u6027\u5316\u97f3\u9891\u5e94\u7528\u7a0b\u5e8f\u3001\u865a\u62df\u52a9\u624b\u7b49\u7684\u53ef\u80fd\u6027\u3002<\/p>\n<hr>\n<p>\n","protected":false},"excerpt":{"rendered":"<p>\u91cd\u8981\u63d0\u793a\uff1a\u672c\u6307\u5357\u4ec5\u4f9b\u7b26\u5408\u9053\u5fb7\u89c4\u8303\u7684\u4f7f\u7528\u3002\u8bf7\u786e\u4fdd\u4f60\u62e5\u6709\u7528\u4e8e\u6570\u636e\u96c6\u521b\u5efa\u7684\u4efb\u4f55\u5a92\u4f53\u7684\u5fc5\u8981\u6743\u9650\uff0c\u5e76\u9075\u5b88\u9002\u7528\u7684\u6cd5\u5f8b\u548c\u51c6\u5219\u3002\u4f7f\u7528 YouTube-DL \u7b49\u5de5\u5177\u65f6\uff0c\u6211\u4eec\u5efa\u8bae\u4e0b\u8f7d\u548c\u5904\u7406\u4f60\u62e5\u6709\u6240\u6709\u6743\u7684\u5185\u5bb9\uff08\u4f8b\u5982\u60a8\u81ea\u5df1\u7684\u5f55\u97f3\uff09\uff0c\u4ee5\u5c0a\u91cd\u4ed6\u4eba\u7684\u77e5\u8bc6\u4ea7\u6743\u3002\u8bed\u97f3\u514b\u9686\u6280\u672f\u5e94\u59cb\u7ec8\u4ee5\u8d1f\u8d23\u4efb\u7684\u65b9\u5f0f\u4f7f\u7528\uff0c\u5e76\u627f\u8bfa\u4fdd\u62a4\u9690\u79c1\u5e76\u5c0a\u91cd\u4e2a\u4eba\u3002 \u968f\u7740\u63d0\u4f9b\u914d\u97f3\u548c\u7c7b\u4f3c\u4eba\u7c7b\u5bf9\u8bdd\u58f0\u97f3\u7684 AI \u5e94\u7528\u7a0b\u5e8f\u7684\u5174\u8d77\uff0c\u4eba\u4eec\u5bf9\u6784\u5efa\u81ea\u5b9a\u4e49\u6587\u672c\u8f6c\u8bed\u97f3\u6a21\u578b\u7684\u5174\u8da3\u65e5\u76ca\u6d53\u539a\u3002\u8bb8\u591a\u5f00\u53d1\u4eba\u5458\u548c\u516c\u53f8\u8bd5\u56fe\u901a\u8fc7\u5fae\u8c03\u81ea\u5df1\u7684\u6a21\u578b\u6765\u907f\u514d\u4ed8\u8d39\u914d\u97f3\u670d\u52a1\u7684\u6210\u672c\u3002\u7136\u800c\uff0c\u521b\u5efa\u9ad8\u8d28\u91cf\u6587\u672c\u8f6c\u8bed\u97f3\u7cfb\u7edf\u7684\u7b2c\u4e00\u6b65\u4e5f\u662f\u6700\u5173\u952e\u7684\u4e00\u6b65\u662f\u83b7\u53d6\u4e30\u5bcc\u3001\u51c6\u5907\u5145\u5206\u7684\u6570\u636e\u96c6\u3002\u672c\u6307\u5357\u4ecb\u7ecd\u4e86\u6784\u5efa\u6b64\u7c7b\u6570\u636e\u96c6\u7684\u5168\u9762\u8fc7\u7a0b\uff0c\u91cd\u70b9\u662f\u63d0\u53d6\u6709\u6548\u8bed\u97f3\u514b\u9686\u6240\u5fc5\u9700\u7684\u6e05\u6670\u51c6\u786e\u7684\u58f0\u97f3\u6837\u672c\u3002 \u4f5c\u4e3a\u6211\u9879\u76ee\u7684\u4e2d\u95f4\u6b65\u9aa4\uff0c\u6211\u6b63\u5728\u52aa\u529b\u5efa\u7acb\u4e00\u4e2a\u53ef\u4ee5\u65e0\u7f1d\u6267\u884c\u6bcf\u4e2a\u6b65\u9aa4\u7684\u81ea\u52a8\u5316\u7ba1\u9053\u3002\u672c\u6307\u5357\u63d0\u4f9b\u4e86\u6709\u5173\u521b\u5efa\u9ad8\u8d28\u91cf\u6570\u636e\u96c6\u7684\u8be6\u7ec6\u6f14\u7ec3\uff0c\u6db5\u76d6\u4e86\u4ece\u89c6\u9891\u4e0b\u8f7d\u5230\u97f3\u9891\u8f6c\u5f55\u7684\u6240\u6709\u5185\u5bb9\u3002\u6700\u540e\uff0c\u6211\u5c06\u8ba8\u8bba\u5728\u6b64\u8fc7\u7a0b\u4e2d\u9047\u5230\u7684\u4e00\u4e9b\u6311\u6218\u3002 \u8bed\u97f3\u63d0\u53d6\u6982\u8ff0 \u6b64\u5de5\u4f5c\u6d41\u7a0b\u4f7f\u7528 YouTube-DL\u3001FFMPEG\u3001WhisperX \u7b49\u5de5\u5177\u6765\u786e\u4fdd\u63d0\u53d6\u7684\u97f3\u9891\u7b26\u5408\u9ad8\u8d28\u91cf\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684\u8981\u6c42\u3002\u8ba9\u6211\u4eec\u6df1\u5165\u4e86\u89e3\u6bcf\u4e2a\u6b65\u9aa4\u4ee5\u53ca\u5e2e\u52a9\u60a8\u5165\u95e8\u7684\u5fc5\u8981\u547d\u4ee4\u3002 1\u3001\u4e0b\u8f7d\u97f3\u9891\u6e90 \u7b2c\u4e00\u6b65\u662f\u4ece CSV \u6587\u4ef6\u4e2d\u63d0\u4f9b\u7684 URL \u5217\u8868\u4e2d\u4e0b\u8f7d\u89c6\u9891\u3002 \u662f\u4e00\u4e2a\u7b80\u5316\u6b64\u8fc7\u7a0b\u7684\u547d\u4ee4\u884c\u5de5\u5177\u3002\u5b83\u5141\u8bb8\u60a8\u4e0b\u8f7d\u9ad8\u8d28\u91cf\u7684\u89c6\u9891\u5e76\u652f\u6301\u5404\u79cd\u89c6\u9891\u6258\u7ba1\u5e73\u53f0\u3002 \u4ece CSV \u52a0\u8f7d URL\uff1a\u4ece CSV \u6587\u4ef6\u4e2d\u63d0\u53d6 URL \u5217\u8868\u3002 \u8fd0\u884c YouTube-DL\uff1a\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u4ee5\u6700\u4f73\u53ef\u7528\u8d28\u91cf\u4e0b\u8f7d\u89c6\u9891\u3002 \u7528\u6cd5\uff1a youtube-dl -f &#8216;bestvideo[ext=mp4]+bestaudio[ext=m4a]\/mp4&#8217; &lt;URL&gt; \u6ce8\u610f\uff1a\u4ece 2024 \u5e74 8 \u6708\u5f00\u59cb\uff0cyoutube-dl \u8981\u6c42\u4f60\u5728\u4ece AWS EC2 \u8fd0\u884c\u65f6\u4f7f\u7528 OAuth\/Cookies \u4f20\u9012\u8eab\u4efd\u9a8c\u8bc1\u8be6\u7ec6\u4fe1\u606f\uff0c\u56e0\u4e3a Google \u5bf9\u4e0b\u8f7d T&amp;C \u975e\u5e38\u4e25\u683c\uff0c\u5e76\u5c06\u8bb8\u591a\u516c\u5171\u4e91\u63d0\u4f9b\u5546 IP \u5217\u5165\u9ed1\u540d\u5355\u3002 2\u3001\u5206\u79bb\u97f3\u89c6\u9891 \u4e0b\u8f7d\u89c6\u9891\u540e\uff0c\u4e0b\u4e00\u6b65\u662f\u5c06\u97f3\u8f68\u4e0e\u89c6\u9891\u6587\u4ef6\u5206\u79bb\u3002 \u662f\u4e00\u4e2a\u529f\u80fd\u5f3a\u5927\u7684\u5de5\u5177\uff0c\u53ef\u4ee5\u8f7b\u677e\u63d0\u53d6\u548c\u8f6c\u6362\u97f3\u9891\u683c\u5f0f\u3002\u5c06\u97f3\u9891\u8f6c\u6362\u4e3a WAV \u683c\u5f0f\u53ef\u786e\u4fdd\u4e0e\u5927\u591a\u6570\u673a\u5668\u5b66\u4e60\u6846\u67b6\u517c\u5bb9\u3002 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[13],"tags":[],"class_list":["post-53760","post","type-post","status-publish","format-standard","hentry","category-ai"],"_links":{"self":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/posts\/53760","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/comments?post=53760"}],"version-history":[{"count":0,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/posts\/53760\/revisions"}],"wp:attachment":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/media?parent=53760"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/categories?post=53760"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/tags?post=53760"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}