{"id":27236,"date":"2025-02-28T19:39:17","date_gmt":"2025-02-28T11:39:17","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=27236"},"modified":"2025-05-22T23:12:39","modified_gmt":"2025-05-22T15:12:39","slug":"baichuan-audio","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/baichuan-audio\/","title":{"rendered":"Baichuan-Audio"},"content":{"rendered":"<p>Baichuan-Audio \u662f\u7531\u767e\u5ddd\u667a\u80fd\uff08baichuan-inc\uff09\u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u6258\u7ba1\u4e8e GitHub \u4e0a\uff0c\u4e13\u6ce8\u4e8e\u7aef\u5230\u7aef\u7684\u8bed\u97f3\u4ea4\u4e92\u6280\u672f\u3002\u8be5\u9879\u76ee\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5b8c\u6574\u7684\u97f3\u9891\u5904\u7406\u6846\u67b6\uff0c\u80fd\u591f\u5c06\u8bed\u97f3\u8f93\u5165\u8f6c\u5316\u4e3a\u79bb\u6563\u97f3\u9891\u6807\u8bb0\uff0c\u518d\u901a\u8fc7\u5927\u6a21\u578b\u751f\u6210\u5bf9\u5e94\u7684\u6587\u672c\u548c\u97f3\u9891\u8f93\u51fa\uff0c\u5b9e\u73b0\u9ad8\u8d28\u91cf\u7684\u5b9e\u65f6\u5bf9\u8bdd\u529f\u80fd\u3002\u652f\u6301\u4e2d\u82f1\u6587\u53cc\u8bed\uff0c\u9002\u7528\u4e8e\u9700\u8981\u8bed\u97f3\u7406\u89e3\u548c\u751f\u6210\u7684\u573a\u666f\uff0c\u4f8b\u5982\u667a\u80fd\u52a9\u624b\u3001\u8bed\u97f3\u804a\u5929\u673a\u5668\u4eba\u7b49\u3002\u6b64\u5916\uff0c\u9879\u76ee\u8fd8\u5f00\u6e90\u4e86\u57fa\u7840\u6a21\u578b Baichuan-Audio-Base \u548c\u8bc4\u6d4b\u57fa\u51c6 OpenAudio-Bench\uff0c\u4e3a\u5f00\u53d1\u8005\u63d0\u4f9b\u5f3a\u5927\u7684\u7814\u7a76\u548c\u5f00\u53d1\u652f\u6301\u3002\u9879\u76ee\u9075\u5faa Apache 2.0 \u8bb8\u53ef\u8bc1\uff0c\u9002\u5408\u5b66\u672f\u7814\u7a76\u548c\u7279\u5b9a\u6761\u4ef6\u4e0b\u7684\u5546\u4e1a\u5e94\u7528\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-27237\" title=\"Baichuan-Audio\uff1a\u652f\u6301\u5b9e\u65f6\u8bed\u97f3\u4ea4\u4e92\u7684\u7aef\u5230\u7aef\u97f3\u9891\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/4190aadd3d1f736.jpg\" alt=\"Baichuan-Audio\uff1a\u652f\u6301\u5b9e\u65f6\u8bed\u97f3\u4ea4\u4e92\u7684\u7aef\u5230\u7aef\u97f3\u9891\u6a21\u578b-1\" width=\"2670\" height=\"1326\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/4190aadd3d1f736.jpg 2670w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/4190aadd3d1f736-768x381.jpg 768w\" sizes=\"auto, (max-width: 2670px) 100vw, 2670px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u8bed\u97f3\u8f6c\u97f3\u9891\u6807\u8bb0<\/strong>\uff1a\u901a\u8fc7 Baichuan-Audio Tokenizer \u5c06\u8f93\u5165\u8bed\u97f3\u8f6c\u5316\u4e3a\u79bb\u6563\u97f3\u9891\u6807\u8bb0\uff0c\u652f\u6301 12.5Hz \u5e27\u7387\u8bbe\u8ba1\uff0c\u786e\u4fdd\u4fe1\u606f\u4fdd\u7559\u3002<\/li>\n<li><strong>\u5b9e\u65f6\u8bed\u97f3\u4ea4\u4e92<\/strong>\uff1a\u7ed3\u5408 Audio LLM\uff0c\u652f\u6301\u4e2d\u82f1\u6587\u53cc\u8bed\u5bf9\u8bdd\uff0c\u53ef\u751f\u6210\u9ad8\u8d28\u91cf\u8bed\u97f3\u548c\u6587\u672c\u54cd\u5e94\u3002<\/li>\n<li><strong>\u97f3\u9891\u751f\u6210<\/strong>\uff1a\u5229\u7528\u6d41\u5339\u914d\u89e3\u7801\u5668\uff08Flow-matching Audio Decoder\uff09\uff0c\u4ece\u97f3\u9891\u6807\u8bb0\u751f\u6210\u9ad8\u4fdd\u771f Mel \u9891\u8c31\u56fe\u548c\u6ce2\u5f62\u3002<\/li>\n<li><strong>\u6a21\u578b\u5f00\u6e90<\/strong>\uff1a\u63d0\u4f9b\u672a\u7ecf\u6307\u4ee4\u5fae\u8c03\u7684\u57fa\u7840\u6a21\u578b Baichuan-Audio-Base\uff0c\u5f00\u53d1\u8005\u53ef\u6839\u636e\u9700\u6c42\u8fdb\u884c\u5b9a\u5236\u5316\u8bad\u7ec3\u3002<\/li>\n<li><strong>\u8bc4\u6d4b\u57fa\u51c6<\/strong>\uff1a\u5305\u542b OpenAudio-Bench\uff0c\u63d0\u4f9b 2701 \u4e2a\u6570\u636e\u70b9\u7684\u97f3\u9891\u7406\u89e3\u548c\u751f\u6210\u80fd\u529b\u8bc4\u4f30\u5de5\u5177\u3002<\/li>\n<li><strong>\u591a\u6a21\u6001\u652f\u6301<\/strong>\uff1a\u652f\u6301\u6587\u672c\u548c\u97f3\u9891\u6df7\u5408\u8f93\u5165\uff0c\u5b9e\u73b0\u6a21\u6001\u95f4\u7684\u65e0\u7f1d\u5207\u6362\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<p>\u8981\u5728\u672c\u5730\u4f7f\u7528 Baichuan-Audio\uff0c\u9700\u8981\u642d\u5efa\u5f00\u53d1\u73af\u5883\u5e76\u4e0b\u8f7d\u76f8\u5173\u6a21\u578b\u6587\u4ef6\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u7684\u5b89\u88c5\u6b65\u9aa4\uff1a<\/p>\n<ol>\n<li><strong>\u73af\u5883\u51c6\u5907<\/strong>\n<ul>\n<li>\u786e\u4fdd\u7cfb\u7edf\u5b89\u88c5\u4e86 Python 3.12 \u548c Conda\u3002<\/li>\n<li>\u521b\u5efa\u5e76\u6fc0\u6d3b\u865a\u62df\u73af\u5883\uff1a\n<pre><code>conda create -n baichuan_audio python=3.12\r\nconda activate baichuan_audio\r\n<\/code><\/pre>\n<\/li>\n<li>\u5b89\u88c5\u5fc5\u8981\u7684\u4f9d\u8d56\u5e93\uff1a\n<pre><code>pip install torch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 --index-url https:\/\/download.pytorch.org\/whl\/cu124\r\npip install -r requirements.txt\r\npip install accelerate flash_attn==2.6.3 speechbrain==1.0.0 deepspeed==0.14.4\r\n<\/code><\/pre>\n<\/li>\n<li>\u5b89\u88c5\u7cfb\u7edf\u5de5\u5177\uff1a\n<pre><code>apt install llvm ffmpeg\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u4e0b\u8f7d\u6a21\u578b<\/strong>\n<ul>\n<li>\u4ece GitHub \u4ed3\u5e93\uff08https:\/\/github.com\/baichuan-inc\/Baichuan-Audio\uff09\u514b\u9686\u4ee3\u7801\uff1a\n<pre><code>git clone https:\/\/github.com\/baichuan-inc\/Baichuan-Audio.git\r\ncd Baichuan-Audio\r\n<\/code><\/pre>\n<\/li>\n<li>\u4fee\u6539\u00a0<code>web_demo\/constants.py<\/code>\u00a0\u4e2d\u7684\u00a0<code>MODEL_PATH<\/code>\uff0c\u6307\u5411\u672c\u5730\u6a21\u578b\u8def\u5f84\uff08\u9700\u624b\u52a8\u4e0b\u8f7d\u6216\u901a\u8fc7 Hugging Face \u83b7\u53d6\u6a21\u578b\u6743\u91cd\uff0c\u4f8b\u5982\u00a0<code>baichuan-inc\/Baichuan-Audio<\/code>\uff09\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u8fd0\u884c Demo<\/strong>\n<ul>\n<li>\u542f\u52a8\u8bed\u97f3\u8bc6\u522b Demo\uff1a\n<pre><code>cd web_demo\r\npython base_asr_demo.py\r\n<\/code><\/pre>\n<\/li>\n<li>\u542f\u52a8\u8bed\u97f3\u5408\u6210 Demo\uff1a\n<pre><code>python base_tts_demo.py\r\n<\/code><\/pre>\n<\/li>\n<li>\u542f\u52a8\u591a\u8f6e\u5bf9\u8bdd Demo\uff1a\n<pre><code>python s2s_gradio_demo_cosy_multiturn.py\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<h3>\u529f\u80fd\u64cd\u4f5c\u6d41\u7a0b<\/h3>\n<h4>1. \u8bed\u97f3\u8f6c\u97f3\u9891\u6807\u8bb0\uff08Baichuan-Audio Tokenizer\uff09<\/h4>\n<ul>\n<li><strong>\u529f\u80fd\u8bf4\u660e<\/strong>\uff1a\u5c06\u8f93\u5165\u7684\u8bed\u97f3\u6587\u4ef6\u6216\u5b9e\u65f6\u9ea6\u514b\u98ce\u8f93\u5165\u8f6c\u5316\u4e3a\u79bb\u6563\u7684\u97f3\u9891\u6807\u8bb0\uff0c\u7528\u4e8e\u540e\u7eed\u5904\u7406\u3002<\/li>\n<li><strong>\u64cd\u4f5c\u6b65\u9aa4<\/strong>\uff1a\n<ol>\n<li>\u51c6\u5907\u97f3\u9891\u6587\u4ef6\uff08\u652f\u6301 WAV \u683c\u5f0f\uff09\u6216\u8fde\u63a5\u9ea6\u514b\u98ce\u8bbe\u5907\u3002<\/li>\n<li>\u5728\u8fd0\u884c\u00a0<code>base_asr_demo.py<\/code>\u00a0\u540e\uff0c\u754c\u9762\u4f1a\u63d0\u793a\u4e0a\u4f20\u97f3\u9891\u6216\u5f00\u542f\u9ea6\u514b\u98ce\u3002<\/li>\n<li>\u7cfb\u7edf\u81ea\u52a8\u8c03\u7528 <a href=\"https:\/\/www.kdjingpai.com\/de\/crisperwhisper\/\">Whisper<\/a> Large Encoder \u63d0\u53d6\u7279\u5f81\uff0c\u5e76\u901a\u8fc7 8 \u5c42 RVQ\uff08\u6b8b\u5dee\u77e2\u91cf\u91cf\u5316\uff09\u751f\u6210\u97f3\u9891\u6807\u8bb0\u3002<\/li>\n<li>\u8f93\u51fa\u7ed3\u679c\u53ef\u5728\u63a7\u5236\u53f0\u67e5\u770b\uff0c\u683c\u5f0f\u4e3a\u79bb\u6563\u6807\u8bb0\u5e8f\u5217\u3002<\/li>\n<\/ol>\n<\/li>\n<li><strong>\u6ce8\u610f\u4e8b\u9879<\/strong>\uff1a\u786e\u4fdd\u97f3\u9891\u6e05\u6670\uff0c\u80cc\u666f\u566a\u97f3\u4f4e\uff0c\u4ee5\u63d0\u9ad8\u8bc6\u522b\u51c6\u786e\u7387\u3002<\/li>\n<\/ul>\n<h4>2. \u5b9e\u65f6\u8bed\u97f3\u4ea4\u4e92\uff08Audio LLM\uff09<\/h4>\n<ul>\n<li><strong>\u529f\u80fd\u8bf4\u660e<\/strong>\uff1a\u652f\u6301\u7528\u6237\u901a\u8fc7\u8bed\u97f3\u6216\u6587\u5b57\u8f93\u5165\uff0c\u4e0e\u6a21\u578b\u8fdb\u884c\u4e2d\u82f1\u6587\u53cc\u8bed\u5b9e\u65f6\u5bf9\u8bdd\u3002<\/li>\n<li><strong>\u64cd\u4f5c\u6b65\u9aa4<\/strong>\uff1a\n<ol>\n<li>\u8fd0\u884c\u00a0<code>s2s_gradio_demo_cosy_multiturn.py<\/code>\uff0c\u6253\u5f00 Gradio \u754c\u9762\u3002<\/li>\n<li>\u70b9\u51fb\u201c\u5f55\u97f3\u201d\u6309\u94ae\uff0c\u5f00\u59cb\u8bed\u97f3\u8f93\u5165\uff08\u4f8b\u5982\uff1a\u201c\u8bf7\u7528\u4e2d\u6587\u544a\u8bc9\u6211\u4eca\u5929\u7684\u5929\u6c14\u201d\uff09\u3002<\/li>\n<li>\u7cfb\u7edf\u5c06\u8bed\u97f3\u8f6c\u4e3a\u6807\u8bb0\uff0cAudio LLM \u751f\u6210\u6587\u672c\u548c\u8bed\u97f3\u54cd\u5e94\u3002<\/li>\n<li>\u54cd\u5e94\u4f1a\u540c\u65f6\u4ee5\u6587\u5b57\u663e\u793a\u5e76\u901a\u8fc7\u626c\u58f0\u5668\u64ad\u653e\u3002<\/li>\n<\/ol>\n<\/li>\n<li><strong>\u7279\u8272\u7528\u6cd5<\/strong>\uff1a\n<ul>\n<li>\u8f93\u5165\u6df7\u5408\u6a21\u6001\u5185\u5bb9\uff0c\u4f8b\u5982\u5148\u8f93\u5165\u6587\u5b57\u201cHi\u201d\uff0c\u518d\u7528\u8bed\u97f3\u8bf4\u201c\u8bf7\u7ee7\u7eed\u7528\u82f1\u8bed\u56de\u7b54\u201d\u3002<\/li>\n<li>\u7cfb\u7edf\u901a\u8fc7\u7279\u6b8a\u6807\u8bb0\u5b9e\u73b0\u6a21\u6001\u5207\u6362\uff0c\u4fdd\u6301\u5bf9\u8bdd\u8fde\u8d2f\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h4>3. \u97f3\u9891\u751f\u6210\uff08Flow-matching Audio Decoder\uff09<\/h4>\n<ul>\n<li><strong>\u529f\u80fd\u8bf4\u660e<\/strong>\uff1a\u6839\u636e\u6587\u672c\u6216\u97f3\u9891\u8f93\u5165\uff0c\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u8bed\u97f3\u8f93\u51fa\u3002<\/li>\n<li><strong>\u64cd\u4f5c\u6b65\u9aa4<\/strong>\uff1a\n<ol>\n<li>\u8fd0\u884c\u00a0<code>base_tts_demo.py<\/code>\uff0c\u8fdb\u5165\u6587\u672c\u5230\u8bed\u97f3\u7684\u754c\u9762\u3002<\/li>\n<li>\u5728\u6587\u672c\u6846\u8f93\u5165\u5185\u5bb9\uff08\u4f8b\u5982\uff1a\u201c\u4f60\u597d\uff0c\u4eca\u5929\u662f\u661f\u671f\u4e94\u201d\uff09\u3002<\/li>\n<li>\u70b9\u51fb\u201c\u751f\u6210\u201d\u6309\u94ae\uff0c\u7cfb\u7edf\u5c06\u6587\u672c\u8f6c\u4e3a\u97f3\u9891\u6807\u8bb0\uff0c\u518d\u901a\u8fc7\u6d41\u5339\u914d\u89e3\u7801\u5668\u751f\u6210 Mel \u9891\u8c31\u56fe\u3002<\/li>\n<li>\u4f7f\u7528\u5185\u7f6e\u58f0\u7801\u5668\u5c06\u9891\u8c31\u56fe\u8f6c\u4e3a WAV \u6587\u4ef6\uff0c\u81ea\u52a8\u64ad\u653e\u6216\u4e0b\u8f7d\u3002<\/li>\n<\/ol>\n<\/li>\n<li><strong>\u9ad8\u7ea7\u529f\u80fd<\/strong>\uff1a\u652f\u6301\u8c03\u6574\u8bed\u901f\u548c\u97f3\u8c03\uff0c\u5177\u4f53\u53c2\u6570\u53ef\u5728\u4ee3\u7801\u4e2d\u4fee\u6539\u3002<\/li>\n<\/ul>\n<h4>4. \u4f7f\u7528\u8bc4\u6d4b\u57fa\u51c6\uff08OpenAudio-Bench\uff09<\/h4>\n<ul>\n<li><strong>\u529f\u80fd\u8bf4\u660e<\/strong>\uff1a\u8bc4\u4f30\u6a21\u578b\u7684\u97f3\u9891\u7406\u89e3\u548c\u751f\u6210\u80fd\u529b\u3002<\/li>\n<li><strong>\u64cd\u4f5c\u6b65\u9aa4<\/strong>\uff1a\n<ol>\n<li>\u4e0b\u8f7d OpenAudio-Bench \u6570\u636e\u96c6\uff08\u4f4d\u4e8e GitHub \u4ed3\u5e93\uff09\u3002<\/li>\n<li>\u5728\u672c\u5730\u8fd0\u884c\u8bc4\u6d4b\u811a\u672c\uff08\u9700\u81ea\u884c\u7f16\u5199\u6216\u53c2\u8003\u4ed3\u5e93\u6587\u6863\uff09\u3002<\/li>\n<li>\u8f93\u5165\u6d4b\u8bd5\u97f3\u9891\u6216\u6587\u672c\uff0c\u83b7\u53d6\u6a21\u578b\u5728 5 \u4e2a\u5b50\u8bc4\u4f30\u96c6\u4e0a\u7684\u8868\u73b0\u8bc4\u5206\u3002<\/li>\n<\/ol>\n<\/li>\n<li><strong>\u5e94\u7528\u573a\u666f<\/strong>\uff1a\u5f00\u53d1\u8005\u53ef\u7528\u4e8e\u5bf9\u6bd4\u4e0d\u540c\u6a21\u578b\u7684\u6027\u80fd\uff0c\u4f18\u5316\u8bad\u7ec3\u7b56\u7565\u3002<\/li>\n<\/ul>\n<h3>\u6ce8\u610f\u4e8b\u9879<\/h3>\n<ul>\n<li><strong>\u786c\u4ef6\u8981\u6c42<\/strong>\uff1a\u5efa\u8bae\u4f7f\u7528\u652f\u6301 CUDA \u7684 GPU\uff08\u5982 NVIDIA \u663e\u5361\uff09\uff0cCPU \u8fd0\u884c\u53ef\u80fd\u8f83\u6162\u3002<\/li>\n<li><strong>\u7f51\u7edc\u4f9d\u8d56<\/strong>\uff1a\u9996\u6b21\u8fd0\u884c\u9700\u8054\u7f51\u4e0b\u8f7d\u6a21\u578b\u6743\u91cd\uff0c\u540e\u7eed\u53ef\u79bb\u7ebf\u4f7f\u7528\u3002<\/li>\n<li><strong>\u5546\u4e1a\u4f7f\u7528<\/strong>\uff1a\u9700\u9075\u5b88 Apache 2.0 \u534f\u8bae\uff0c\u5e76\u786e\u8ba4\u65e5\u6d3b\u8dc3\u7528\u6237\uff08DAU\uff09\u4f4e\u4e8e 100 \u4e07\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Baichuan-Audio \u662f\u7531\u767e\u5ddd\u667a\u80fd\uff08baichuan-inc\uff09\u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u6258\u7ba1\u4e8e GitHub \u4e0a\uff0c\u4e13\u6ce8\u4e8e\u7aef\u5230\u7aef\u7684\u8bed\u97f3\u4ea4\u4e92\u6280\u672f\u3002\u8be5\u9879\u76ee\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5b8c\u6574\u7684\u97f3\u9891\u5904\u7406\u6846\u67b6\uff0c\u80fd\u591f\u5c06\u8bed\u97f3\u8f93\u5165\u8f6c\u5316\u4e3a\u79bb\u6563\u97f3\u9891\u6807\u8bb0\uff0c\u518d\u901a\u8fc7\u5927\u6a21\u578b\u751f\u6210\u5bf9\u5e94\u7684\u6587\u672c&#8230;<\/p>\n","protected":false},"author":1,"featured_media":61938,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[392,400],"tags":[230,374],"class_list":["post-27236","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-models","category-speech-model","tag-aikaiyuanxiangmu","tag-duandaoduanyuyinzhubei"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/27236","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/comments?post=27236"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/27236\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media\/61938"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media?parent=27236"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/categories?post=27236"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/tags?post=27236"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}