
{"id":13749,"date":"2026-10-22T20:39:29","date_gmt":"2026-10-22T12:39:29","guid":{"rendered":"https:\/\/infernews.com\/blog\/rubatocodec-2\/"},"modified":"2026-10-22T21:32:12","modified_gmt":"2026-10-22T13:32:12","slug":"rubatocodec-2","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/rubatocodec-2\/","title":{"rendered":"RubatoCodec\uff1a\u6df1\u5165\u795e\u7d93\u97f3\u8a0a\u7de8\u78bc &#8211; \u6b98\u5dee\u5411\u91cf\u91cf\u5316\uff08RVQ\uff09\u8207\u9ad8\u4fdd\u771f\u91cd\u69cb\u67b6\u69cb"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/10\/RubatoCodec-RVQ.jpg\" alt=\"RubatoCodec architecture\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u8a9e\u97f3\u7247\u6bb5\u7684\u8cc7\u8a0a\u5bc6\u5ea6\u5404\u6709\u4e0d\u540c\uff0c\u82e5\u4e00\u5f8b\u6309\u56fa\u5b9a\u9577\u5ea6\u5207\u5206\uff0c\u4fbf\u96e3\u4ee5\u6309\u5167\u5bb9\u5206\u914d\u7de8\u78bc\u8cc7\u6e90\u3002RubatoCodec \u662f\u795e\u7d93\u8a9e\u97f3\u7de8\u78bc\u5668\uff0c\u6703\u628a\u8a9e\u97f3\u8f49\u6210\u5e36\u6709\u96e2\u6563\u78bc\u7d22\u5f15\u548c\u6301\u7e8c\u6642\u9593\u7684\u53ef\u8b8a\u9577\u4e8b\u4ef6\uff0c\u4e26\u652f\u63f4\u9010\u6bb5\u91cd\u5efa\u6ce2\u5f62\uff0c\u9069\u5408\u9700\u8981\u63a7\u5236\u78bc\u7387\u6216\u4e32\u6d41\u8f38\u51fa\u7684\u8a9e\u97f3\u7cfb\u7d71\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u5728 50 Hz \u6f5b\u5728\u8868\u793a\u4e0a\uff0c\u53c3\u8003\u591a\u5c3a\u5ea6\u6f5b\u5728\u7279\u5fb5\u5dee\u7570\u548c\u6ce2\u5f62\u983b\u8b5c\u901a\u91cf\uff0c\u7d2f\u7a4d\u5206\u914d\u4e8b\u4ef6\uff1b\u9810\u8a2d\u76ee\u6a19\u6bd4\u4f8b\u70ba 0.25\uff0c\u6574\u6bb5\u8a9e\u97f3\u7684\u4e8b\u4ef6\u6578\u5247\u6309\u516c\u5f0f\u7cbe\u78ba\u63a7\u5236\u3002\u9019\u7a2e\u505a\u6cd5\u6bcb\u9808\u53e6\u8a2d\u908a\u754c\u9810\u6e2c\u5668\u3001\u908a\u754c\u6a19\u7c64\u6216 rate loss\uff0c\u4f46\u4ecd\u9700\u628a\u4e8b\u4ef6\u6301\u7e8c\u6642\u9593\u4e00\u4f75\u7d0d\u5165\u7de8\u78bc\uff0c\u4e0d\u80fd\u53ea\u770b\u78bc\u7d22\u5f15\u7684\u4f4d\u5143\u7387\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img data-dominant-color=\"e8e8e3\" data-has-transparency=\"false\" style=\"--dominant-color: #e8e8e3;\" loading=\"lazy\" decoding=\"async\" width=\"800\" height=\"447\" sizes=\"auto, (max-width: 800px) 100vw, 800px\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/10\/RubatoCodec-1.jpg\" alt=\"\" class=\"wp-image-13755 not-transparent\" srcset=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/10\/RubatoCodec-1.jpg 800w, https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/10\/RubatoCodec-1-768x429.jpg 768w, https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/10\/RubatoCodec-1-300x168.jpg 300w\" \/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u91cf\u5316\u5668\u63a1\u7528 16 \u5c64\u6df7\u5408\u5f0f\u6b98\u5dee\u7d50\u69cb\uff1a\u9996\u5c64\u662f\u63a5\u53d7\u8a9e\u7fa9\u76e3\u7763\u7684 FSQ\uff0c\u5176\u5f8c 15 \u5c64\u4f7f\u7528 DAC-style residual VQ\u3002\u89e3\u78bc\u6642\u53ef\u9078\u7528 1 \u81f3 16 \u5c64\u7684\u4efb\u610f\u524d\u7db4\u6df1\u5ea6\uff0c\u540d\u7fa9\u78bc\u7d22\u5f15\u4f4d\u5143\u7387\u7d04\u70ba 150 \u81f3 2,400 bit\/s\uff0c\u53e6\u52a0\u6301\u7e8c\u6642\u9593\u8cc7\u8a0a\uff1b\u6a21\u578b\u4ee5\u7d04\u6bcf\u79d2 12.5 \u500b\u4e8b\u4ef6\u8868\u793a\u8a9e\u97f3\uff0c\u8f38\u5165\u70ba 16 kHz\uff0c\u56e0\u679c\u4e32\u6d41\u89e3\u78bc\u5668\u8f38\u51fa 44.1 kHz\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6bcf\u500b\u4e8b\u4ef6\u4ee3\u8868\u7d04 80 \u6beb\u79d2\u97f3\u8a0a\uff0c\u5728 H200 \u4e0a\u89e3\u78bc\u7d04\u9700 11 \u6beb\u79d2\uff0c\u5177\u5099\u9010\u4e8b\u4ef6\u6216\u5206\u6279\u89e3\u78bc\u7684\u689d\u4ef6\u3002\u9019\u662f\u786c\u4ef6\u8207\u6a21\u578b\u8a2d\u5b9a\u4e0b\u7684\u901f\u5ea6\u6578\u64da\uff0c\u4e0d\u80fd\u76f4\u63a5\u7576\u6210\u5176\u4ed6\u8a2d\u5099\u4e0a\u7684\u5ef6\u9072\u4fdd\u8b49\uff1b\u73fe\u6709\u8cc7\u8a0a\u4ea6\u672a\u63d0\u4f9b\u8db3\u4ee5\u6bd4\u8f03\u97f3\u8cea\u53d6\u6368\u7684\u5177\u9ad4\u8a55\u6e2c\u7d50\u679c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7814\u7a76\u4eba\u54e1\u3001\u8a9e\u97f3\u6a21\u578b\u958b\u767c\u5718\u968a\uff0c\u4ee5\u53ca\u9700\u8981\u4f4e\u78bc\u7387\u5132\u5b58\u6216\u5373\u6642\u97f3\u8a0a\u8655\u7406\u7684\u7cfb\u7d71\uff0c\u53ef\u7528\u516c\u958b\u6a21\u578b\u548c Python API \u8f09\u5165\u8a9e\u97f3\u3001\u7de8\u78bc\uff0c\u518d\u6309\u6307\u5b9a\u6df1\u5ea6\u89e3\u78bc\u3002\u91cd\u9ede\u5305\u62ec\uff1a<br>\n&#8211; \u4e8b\u4ef6\u9577\u5ea6\u6309\u8a9e\u97f3\u8a0a\u865f\u8b8a\u5316\u8abf\u6574\uff0c\u6574\u6bb5\u4e8b\u4ef6\u6578\u53d7\u76ee\u6a19\u6bd4\u4f8b\u63a7\u5236\u3002<br>\n&#8211; \u652f\u63f4\u9010\u4e8b\u4ef6\u6216\u5206\u6279\u4e32\u6d41\u89e3\u78bc\u3002<br>\n&#8211; \u53ef\u9078\u4e0d\u540c\u91cf\u5316\u5668\u6df1\u5ea6\uff0c\u5728\u78bc\u7387\u8207\u8868\u793a\u5bb9\u91cf\u4e4b\u9593\u8abf\u6574\u3002<br>\n&#8211; \u7de8\u78bc\u8cc7\u8a0a\u9664\u96e2\u6563\u78bc\u7d22\u5f15\u5916\uff0c\u4ea6\u5305\u62ec\u4e8b\u4ef6\u6301\u7e8c\u6642\u9593\u3002<br>\n&#8211; H200 \u4e0a\u6bcf\u4e8b\u4ef6\u7d04\u9700 11 \u6beb\u79d2\u89e3\u78bc\uff0c\u5176\u4ed6\u786c\u4ef6\u8868\u73fe\u672a\u6709\u63d0\u4f9b\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u57fa\u672c\u4f7f\u7528\u7bc4\u4f8b\uff08\u63a8\u8ad6\u8207\u91cd\u69cb\u6d41\u7a0b\uff09<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u5c08\u6848\u901a\u5e38\u9075\u5faa\u6a19\u6e96 PyTorch \u795e\u7d93\u7de8\u89e3\u78bc\u4ecb\u9762\uff0c\u57fa\u672c Python \u8655\u7406\u6d41\u7a0b\u5982\u4e0b\uff1a<\/p>\n\n\n\n<div class=\"codecopy-container\">\n                <div class=\"codecopy-header\">\n                    \n                    <button class=\"codecopy-btn\" data-target=\"codecopy-6adb24f616446\" title=\"Copy code to clipboard\">\n                <svg width=\"16\" height=\"16\" viewBox=\"0 0 24 24\" fill=\"none\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\">\n                    <path d=\"M16 1H4C2.9 1 2 1.9 2 3V17H4V3H16V1ZM19 5H8C6.9 5 6 5.9 6 7V21C6 22.1 6.9 23 8 23H19C20.1 23 21 22.1 21 21V7C21 5.9 20.1 5 19 5ZM19 21H8V7H19V21Z\" fill=\"currentColor\"\/>\n                <\/svg>\n                <span class=\"codecopy-text\">Copy<\/span>\n            <\/button>\n                <\/div>\n                <pre class=\"line-numbers\" id=\"codecopy-6adb24f616446\"><code class=\"language-text\">import torch\nimport torchaudio\nfrom rubatocodec import RubatoCodec\n\n# 1. \u8f09\u5165\u6a21\u578b\u8207\u6b0a\u91cd\ndevice = \"cuda\" if torch.cuda.is_available() else \"cpu\"\nmodel = RubatoCodec.from_pretrained(\"third-intelligence\/RubatoCodec\").to(device)\nmodel.eval()\n\n# 2. \u8b80\u53d6\u4e26\u9810\u8655\u7406\u97f3\u8a0a\uff08\u91cd\u53d6\u6a23\u81f3\u6a21\u578b\u652f\u63f4\u4e4b\u63a1\u6a23\u7387\uff09\nwaveform, sample_rate = torchaudio.load(\"input.wav\")\nif sample_rate != model.sample_rate:\n    resampler = torchaudio.transforms.Resample(sample_rate, model.sample_rate)\n    waveform = resampler(waveform)\n\nwaveform = waveform.unsqueeze(0).to(device)  # &#91;Batch, Channels, Time]\n\n# 3. \u7de8\u78bc\uff1a\u63d0\u53d6\u96e2\u6563 Tokens (RVQ Codes)\nwith torch.no_grad():\n    # tokens \u5f62\u72c0\u901a\u5e38\u70ba &#91;Batch, Num_Quantizers, Frames]\n    tokens = model.encode(waveform)\n\n# 4. \u89e3\u78bc\uff1a\u7531\u96e2\u6563 Tokens \u9084\u539f\u97f3\u8a0a\u6ce2\u5f62\nwith torch.no_grad():\n    reconstructed_waveform = model.decode(tokens)\n\n# 5. \u5132\u5b58\u91cd\u69cb\u97f3\u8a0a\ntorchaudio.save(\"reconstructed.wav\", reconstructed_waveform.squeeze(0).cpu(), model.sample_rate)<\/code><\/pre>\n            <\/div>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/github.com\/third-intelligence\/RubatoCodec\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/huggingface.co\/third-intelligence\/RubatoCodec\" rel=\"noopener noreferrer\"><strong>\u6a21\u578b<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u56fa\u5b9a\u9577\u5ea6\u5207\u5206\u672a\u5fc5\u9069\u5408\u8a9e\u97f3\u7684\u6bcf\u500b\u7247\u6bb5\uff1bRubatoCodec \u6309\u8a0a\u865f\u8b8a\u5316\u5206\u914d\u4e8b\u4ef6\uff0c\u4e26\u53ef\u9010\u6bb5\u89e3\u78bc\uff0c\u9762\u5411\u4f4e\u4f4d\u5143\u7387\u8a9e\u97f3\u8655\u7406\u8207\u4e32\u6d41\u751f\u6210\u3002<\/p>\n","protected":false},"author":8,"featured_media":13748,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","wpai_meta_description":"","footnotes":""},"categories":[133,76,179,137,29,152,128,199],"tags":[],"class_list":["post-13749","post","type-post","status-publish","format-standard","hentry","category-133","category-76","category-nvidia","category-api","category-audio","category-python","category-128","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13749","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=13749"}],"version-history":[{"count":2,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13749\/revisions"}],"predecessor-version":[{"id":13756,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/13749\/revisions\/13756"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/13748"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=13749"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=13749"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=13749"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}