{"id":1155,"date":"2026-06-19T01:22:29","date_gmt":"2026-06-18T17:22:29","guid":{"rendered":"https:\/\/aikuto.com\/?p=1155"},"modified":"2026-06-24T16:04:37","modified_gmt":"2026-06-24T08:04:37","slug":"whisper-transcribe-audio-documents","status":"publish","type":"post","link":"https:\/\/aikuto.com\/?p=1155","title":{"rendered":"whisper\u8f6c\u5f55\u97f3\u9891\u6587\u6863"},"content":{"rendered":"\n

Whisper \u662f OpenAI \u5f00\u53d1\u7684\u4e00\u4e2a\u81ea\u52a8\u8bed\u97f3\u8bc6\u522b\uff08ASR\uff09\u7cfb\u7edf\uff0c\u5b83\u63d0\u4f9b\u4e86\u4e0d\u540c\u5927\u5c0f\u7684\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u6a21\u578b\u8d8a\u5927\uff0c\u901a\u5e38\u5728\u8f6c\u5f55\u4efb\u52a1\u4e2d\u7684\u51c6\u786e\u6027\u5c31\u8d8a\u9ad8\uff0c\u4f46\u540c\u65f6\u4e5f\u9700\u8981\u66f4\u591a\u7684\u8ba1\u7b97\u8d44\u6e90\u548c\u66f4\u957f\u7684\u63a8\u7406\u65f6\u95f4\u3002<\/p>\n\n\n\n

\u6a21\u578b\u5927\u5c0f\u4e0e\u7cbe\u5ea6<\/strong><\/h3>\n\n\n\n

Whisper \u4e00\u5171\u6709 5 \u79cd\u4e0d\u540c\u5927\u5c0f\u7684\u6a21\u578b\uff0c\u6309\u7167\u53c2\u6570\u91cf\u4ece\u5c0f\u5230\u5927\u4f9d\u6b21\u4e3a\u00a0tiny<\/strong><\/mark><\/code>\u3001base<\/strong><\/mark><\/code>\u3001small<\/strong><\/mark><\/code>\u3001medium<\/strong><\/mark><\/code>\u00a0\u548c\u00a0large<\/strong><\/mark><\/code>\uff0c\u5176\u4e2d\u00a0large<\/strong><\/mark><\/code>\u00a0\u53c8\u5206\u4e3a\u00a0large-v1<\/mark><\/strong><\/code>\u00a0\u548c\u00a0large-v2<\/strong><\/code><\/mark>\uff0c<\/mark>large-v2<\/strong><\/code>\u00a0<\/mark>\u662f\u76ee\u524d\u6700\u65b0\u4e14\u8868\u73b0\u6700\u4f18\u7684\u7248\u672c\u3002\u7cbe\u5ea6\u4e0a\uff0ctiny<\/strong><\/mark><\/code>\u00a0\u6a21\u578b\u7684\u51c6\u786e\u6027\u6700\u4f4e\uff0c\u800c\u00a0large-v2<\/code>\u00a0<\/strong><\/mark>\u6a21\u578b\u7684\u51c6\u786e\u6027\u6700\u9ad8\u3002\u4ee5\u4e0b\u662f\u5bf9\u5404\u4e2a\u6a21\u578b\u7684\u7b80\u8981\u4ecb\u7ecd\uff1a<\/p>\n\n\n\n

\u6a21\u578b\u540d\u79f0<\/strong><\/td>\u53c2\u6570\u91cf<\/strong><\/td>\u82f1\u6587\u8f6c\u5f55\u901f\u5ea6<\/strong><\/td>\u591a\u8bed\u8a00\u8f6c\u5f55\u901f\u5ea6<\/strong><\/td>\u7279\u70b9<\/strong><\/td><\/tr>
tiny<\/code><\/td>39M<\/td>~32x<\/td>~16x<\/td>\u901f\u5ea6\u6781\u5feb\uff0c\u4f46\u7cbe\u5ea6\u8f83\u4f4e\uff0c\u9002\u7528\u4e8e\u5bf9\u901f\u5ea6\u8981\u6c42\u6781\u9ad8\u4e14\u5bf9\u8f6c\u5f55\u7cbe\u5ea6\u8981\u6c42\u4e0d\u9ad8\u7684\u573a\u666f\u3002<\/td><\/tr>
base<\/code><\/td>74M<\/td>~16x<\/td>~8x<\/td>\u901f\u5ea6\u8f83\u5feb\uff0c\u7cbe\u5ea6\u4e00\u822c\uff0c\u80fd\u6ee1\u8db3\u4e00\u4e9b\u57fa\u672c\u7684\u8f6c\u5f55\u9700\u6c42\u3002<\/td><\/tr>
small<\/code><\/td>244M<\/td>~6x<\/td>~3x<\/td>\u901f\u5ea6\u9002\u4e2d\uff0c\u7cbe\u5ea6\u8f83\u597d\uff0c\u5728\u5927\u591a\u6570\u60c5\u51b5\u4e0b\u80fd\u63d0\u4f9b\u8f83\u4e3a\u51c6\u786e\u7684\u8f6c\u5f55\u7ed3\u679c\u3002<\/td><\/tr>
medium<\/code><\/td>769M<\/td>~2x<\/td>~1.5x<\/td>\u901f\u5ea6\u76f8\u5bf9\u8f83\u6162\uff0c\u4f46\u7cbe\u5ea6\u8f83\u9ad8\uff0c\u80fd\u5904\u7406\u4e00\u4e9b\u590d\u6742\u7684\u8bed\u97f3\u5185\u5bb9\u3002<\/td><\/tr>
large-v1<\/code><\/td>1550M<\/td>1x<\/td>1x<\/td>\u7cbe\u5ea6\u5f88\u9ad8\uff0c\u4f46\u901f\u5ea6\u6700\u6162\uff0c\u5bf9\u8ba1\u7b97\u8d44\u6e90\u8981\u6c42\u9ad8\u3002<\/td><\/tr>
large-v2<\/code><\/td>1550M<\/td>1x<\/td>1x<\/td>\u5728 large-v1<\/code> \u7684\u57fa\u7840\u4e0a\u8fdb\u4e00\u6b65\u4f18\u5316\uff0c\u662f\u76ee\u524d Whisper \u6a21\u578b\u4e2d\u7cbe\u5ea6\u6700\u9ad8\u7684\u7248\u672c\u3002<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

\u793a\u4f8b\u4ee3\u7801<\/strong><\/h3>\n\n\n\n

\u5982\u679c\u4f60\u60f3\u4f7f\u7528 large-v2<\/code> \u6a21\u578b\u8fdb\u884c\u8f6c\u5f55\uff0c\u53ef\u4ee5\u5c06\u4ee3\u7801\u4e2d\u7684 model_name<\/code> \u53d8\u91cf\u4fee\u6539\u4e3a \"large-v2\"<\/code>\uff1a<\/p>\n\n\n

import whisper\n\n# \u9009\u62e9 large-v2 \u6a21\u578b<\/span>\nmodel_name = \"large-v2\"<\/span>\ndevice = \"cuda\"<\/span> if<\/span> torch.cuda.is_available() else<\/span> \"cpu\"<\/span>\n\n# \u52a0\u8f7d\u6a21\u578b<\/span>\nmodel = whisper.load_model(model_name, device=device)\n\n# \u8fdb\u884c\u8f6c\u5f55<\/span>\naudio_file = \"your_audio_file.mp3\"<\/span>\nresult = model.transcribe(audio_file, language=\"Japanese\"<\/span>, word_timestamps=True<\/span>)\n\n# \u8f93\u51fa\u8f6c\u5f55\u7ed3\u679c<\/span>\nprint<\/span>(result[\"text\"<\/span>])<\/code><\/span>Code language:<\/span> PHP<\/span> (<\/span>php<\/span>)<\/span><\/small><\/pre>\n\n\n
\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u4f60\u9700\u8981\u6839\u636e\u81ea\u5df1\u7684\u8ba1\u7b97\u8d44\u6e90\u3001\u65f6\u95f4\u9650\u5236\u548c\u5bf9\u8f6c\u5f55\u7cbe\u5ea6\u7684\u8981\u6c42\u6765\u9009\u62e9\u5408\u9002\u7684\u6a21\u578b\u3002\u5982\u679c\u4f60\u7684\u8ba1\u7b97\u8d44\u6e90\u5145\u8db3\uff0c\u4e14\u5bf9\u8f6c\u5f55\u7cbe\u5ea6\u6709\u8f83\u9ad8\u8981\u6c42\uff0c\u90a3\u4e48\u53ef\u4ee5\u8003\u8651\u4f7f\u7528 large-v2<\/code> \u6a21\u578b\uff1b\u5982\u679c\u5bf9\u901f\u5ea6\u8981\u6c42\u8f83\u9ad8\uff0cmedium<\/code> \u6216\u66f4\u5c0f\u7684\u6a21\u578b\u53ef\u80fd\u66f4\u9002\u5408\u4f60\u3002<\/p>\n\n\n\n
\u4e00\u3001 Python \u7248\u672c\u95ee\u9898<\/strong><\/p>\n\n\n\n
openai-whisper<\/code> \u4e0d\u652f\u6301 Python 3.6\uff0c\u4f60\u9700\u8981\u5c06 Python \u7248\u672c\u5347\u7ea7\u5230 3.7 \u53ca\u4ee5\u4e0a\u3002\u5728 CentOS \u7cfb\u7edf\u4e0a\uff0c\u4f60\u53ef\u4ee5\u6309\u7167\u4ee5\u4e0b\u6b65\u9aa4\u5347\u7ea7 Python\uff1a<\/p>\n\n\n\n
1.1 \u5b89\u88c5\u4f9d\u8d56<\/strong><\/h3>\n\n\n
sudo yum install -y gcc openssl-devel bzip2-devel libffi-devel zlib-devel<\/code><\/span><\/pre>\n\n\n
1.2 \u4e0b\u8f7d Python 3.8\uff08\u8fd9\u91cc\u4ee5 3.8 \u4e3a\u4f8b\uff09<\/strong><\/h3>\n\n\n
wget https:\/\/www.python.org\/ftp\/python\/3.8.16\/Python-3.8.16.tgz<\/span><\/code><\/span>Code language:<\/span> JavaScript<\/span> (<\/span>javascript<\/span>)<\/span><\/small><\/pre>\n\n\n
1.3 \u89e3\u538b\u5e76\u7f16\u8bd1\u5b89\u88c5<\/strong><\/h3>\n\n\n
tar -xzf Python-3.8<\/span>.16<\/span>.tgz\ncd Python-3.8<\/span>.16<\/span>\n.\/configure --enable-optimizations ## \u5982\u679c\u62a5\u9519\u5c31\u628a--enable-optimizations\u53bb\u6389<\/span>\nmake -j$(nproc)\nsudo make altinstall<\/code><\/span>Code language:<\/span> PHP<\/span> (<\/span>php<\/span>)<\/span><\/small><\/pre>\n\n\n
1.4 \u9a8c\u8bc1\u5b89\u88c5<\/strong><\/h3>\n\n\n
python3<\/span>.8<\/span> --version<\/span><\/code><\/span>Code language:<\/span> CSS<\/span> (<\/span>css<\/span>)<\/span><\/small><\/pre>\n\n\n
2. \u4f7f\u7528\u65b0\u7684 Python \u7248\u672c\u5b89\u88c5 <\/strong>openai-whisper<\/strong><\/code><\/h3>\n\n\n\n
\u5347\u7ea7 Python \u540e\uff0c\u4f7f\u7528\u65b0\u7684 Python \u7248\u672c\u5bf9\u5e94\u7684 pip<\/code> \u6765\u5b89\u88c5 openai-whisper<\/code>\uff1a<\/p>\n\n\n
python3.8<\/span> -m pip install --user git+https:\/\/github.com\/openai\/whisper.git<\/span><\/code><\/span>Code language:<\/span> JavaScript<\/span> (<\/span>javascript<\/span>)<\/span><\/small><\/pre>\n\n\n
3. \u5176\u4ed6\u53ef\u80fd\u7684\u95ee\u9898\u53ca\u89e3\u51b3\u529e\u6cd5<\/strong><\/h3>\n\n\n\n
3.1 \u955c\u50cf\u6e90\u95ee\u9898<\/strong><\/h3>\n\n\n\n
\u4f60\u4f7f\u7528\u7684\u662f\u963f\u91cc\u4e91\u7684\u955c\u50cf\u6e90\uff0c\u6709\u65f6\u955c\u50cf\u6e90\u53ef\u80fd\u540c\u6b65\u4e0d\u53ca\u65f6\u6216\u8005\u5b58\u5728\u5176\u4ed6\u95ee\u9898\u3002\u4f60\u53ef\u4ee5\u5c1d\u8bd5\u5207\u6362\u56de\u5b98\u65b9\u7684 PyPI \u6e90\uff1a<\/p>\n\n\n
python3.8<\/span> -m pip install --user -i https:\/\/pypi.org\/simple git+https:\/\/github.com\/openai\/whisper.git<\/span><\/code><\/span>Code language:<\/span> JavaScript<\/span> (<\/span>javascript<\/span>)<\/span><\/small><\/pre>\n\n\n
3.2 \u6743\u9650\u95ee\u9898<\/strong><\/h3>\n\n\n\n
\u786e\u4fdd\u4f60\u6709\u8db3\u591f\u7684\u6743\u9650\u5728\u7528\u6237\u76ee\u5f55\u4e0b\u5b89\u88c5\u5305\u3002\u5982\u679c\u8fd8\u662f\u9047\u5230\u6743\u9650\u95ee\u9898\uff0c\u53ef\u4ee5\u5c1d\u8bd5\u4e34\u65f6\u4f7f\u7528 sudo<\/code> \u6765\u5b89\u88c5\uff08\u4f46\u8981\u8c28\u614e\u4f7f\u7528\uff09\uff1a<\/p>\n\n\n
sudo python3.8<\/span> -m pip install git+https:\/\/github.com\/openai\/whisper.git<\/span><\/code><\/span>Code language:<\/span> JavaScript<\/span> (<\/span>javascript<\/span>)<\/span><\/small><\/pre>\n\n\n
\u901a\u8fc7\u4ee5\u4e0a\u6b65\u9aa4\uff0c\u4f60\u5e94\u8be5\u80fd\u591f\u89e3\u51b3\u5b89\u88c5 openai-whisper<\/code> \u65f6\u9047\u5230\u7684\u95ee\u9898\u3002<\/p>\n\n\n\n
\u591a\u4e2a\u811a\u672c\u88ab\u5b89\u88c5\u5230\u4e86 \/root\/.local\/bin<\/code> \u76ee\u5f55\u4e0b\uff0c\u4f46\u8be5\u76ee\u5f55\u4e0d\u5728\u7cfb\u7edf\u7684 PATH<\/code> \u73af\u5883\u53d8\u91cf\u4e2d\u3002\u8fd9\u610f\u5473\u7740\u4f60\u65e0\u6cd5\u76f4\u63a5\u5728\u547d\u4ee4\u884c\u4e2d\u8c03\u7528\u8fd9\u4e9b\u811a\u672c\uff0c\u9700\u8981\u624b\u52a8\u6dfb\u52a0\u8be5\u76ee\u5f55\u5230 PATH<\/code> \u73af\u5883\u53d8\u91cf\u6216\u8005\u4f7f\u7528\u5b8c\u6574\u7684\u8def\u5f84\u6765\u8c03\u7528\u8fd9\u4e9b\u811a\u672c\u3002\u4ee5\u4e0b\u4e3a\u4f60\u63d0\u4f9b\u51e0\u79cd\u89e3\u51b3\u65b9\u6cd5\uff1a<\/p>\n\n\n\n
\u65b9\u6cd5\u4e00\uff1a\u4e34\u65f6\u6dfb\u52a0\u76ee\u5f55\u5230 <\/strong>PATH<\/strong><\/code><\/h3>\n\n\n\n
\u5982\u679c\u4f60\u53ea\u662f\u60f3\u4e34\u65f6\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u5728\u5f53\u524d\u7ec8\u7aef\u4f1a\u8bdd\u4e2d\u8ba9\u8fd9\u4e9b\u811a\u672c\u53ef\u4ee5\u88ab\u8c03\u7528\uff0c\u53ef\u4ee5\u5728\u7ec8\u7aef\u4e2d\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\uff1a<\/p>\n\n\n
export<\/span> PATH=$PATH:\/root\/<\/span>.local\/bin<\/code><\/span>Code language:<\/span> JavaScript<\/span> (<\/span>javascript<\/span>)<\/span><\/small><\/pre>\n\n\n
\u6267\u884c\u8be5\u547d\u4ee4\u540e\uff0c\u5728\u5f53\u524d\u7ec8\u7aef\u7a97\u53e3\u4e2d\u5c31\u53ef\u4ee5\u76f4\u63a5\u4f7f\u7528\u8fd9\u4e9b\u811a\u672c\u4e86\u3002\u4e0d\u8fc7\uff0c\u8fd9\u79cd\u65b9\u6cd5\u53ea\u5728\u5f53\u524d\u7ec8\u7aef\u4f1a\u8bdd\u4e2d\u6709\u6548\uff0c\u5173\u95ed\u7ec8\u7aef\u540e\u8bbe\u7f6e\u4f1a\u5931\u6548\u3002<\/p>\n\n\n\n
\u65b9\u6cd5\u4e8c\uff1a\u6c38\u4e45\u6dfb\u52a0\u76ee\u5f55\u5230 <\/strong>PATH<\/strong><\/code><\/h3>\n\n\n\n
\u82e5\u4f60\u5e0c\u671b\u6bcf\u6b21\u6253\u5f00\u7ec8\u7aef\u65f6\u8be5\u76ee\u5f55\u90fd\u81ea\u52a8\u6dfb\u52a0\u5230 PATH<\/code> \u4e2d\uff0c\u53ef\u4ee5\u4fee\u6539\u7528\u6237\u7684\u73af\u5883\u53d8\u91cf\u914d\u7f6e\u6587\u4ef6\u3002\u5bf9\u4e8e root<\/code> \u7528\u6237\uff0c\u901a\u5e38\u662f\u4fee\u6539 ~\/.bashrc<\/code> \u6216 ~\/.bash_profile<\/code> \u6587\u4ef6\u3002<\/p>\n\n\n\n
\u4fee\u6539 <\/strong>~\/.bashrc<\/strong><\/code> \u6587\u4ef6<\/strong><\/h3>\n\n\n
nano ~\/.bashrc<\/code><\/span><\/pre>\n\n\n
\u5728\u6587\u4ef6\u672b\u5c3e\u6dfb\u52a0\u4ee5\u4e0b\u5185\u5bb9\uff1a<\/p>\n\n\n
export<\/span> PATH=$PATH:\/root\/<\/span>.local\/bin<\/code><\/span>Code language:<\/span> JavaScript<\/span> (<\/span>javascript<\/span>)<\/span><\/small><\/pre>\n\n\n
\u4fdd\u5b58\u5e76\u9000\u51fa\u6587\u4ef6\uff08\u6309 Ctrl + X<\/code>\uff0c\u7136\u540e\u6309 Y<\/code> \u786e\u8ba4\u4fdd\u5b58\uff0c\u6700\u540e\u6309 Enter<\/code>\uff09\u3002\u63a5\u7740\uff0c\u4f7f\u914d\u7f6e\u6587\u4ef6\u751f\u6548\uff1a<\/p>\n\n\n
source ~\/.bashrc<\/code><\/span><\/pre>\n\n\n
\u4fee\u6539 <\/strong>~\/.bash_profile<\/strong><\/code> \u6587\u4ef6\uff08\u5982\u679c\u4f7f\u7528\u7684\u662f\u8fd9\u4e2a\u6587\u4ef6\uff09<\/strong><\/h3>\n\n\n
nano ~\/.bash_profile<\/code><\/span><\/pre>\n\n\n
\u540c\u6837\u5728\u6587\u4ef6\u672b\u5c3e\u6dfb\u52a0 export PATH=$PATH:\/root\/.local\/bin<\/code>\uff0c\u4fdd\u5b58\u5e76\u9000\u51fa\u540e\u6267\u884c source ~\/.bash_profile<\/code> \u4f7f\u914d\u7f6e\u751f\u6548\u3002<\/p>\n\n\n\n
\u4f7f\u7528 Python 3.8 \u5bf9\u5e94\u7684 <\/strong>pip<\/strong><\/code> \u8fdb\u884c\u5b89\u88c5<\/strong><\/h3>\n\n\n\n
\u4f7f\u7528 python3.8 -m pip<\/code> \u6765\u786e\u4fdd\u5b89\u88c5\u64cd\u4f5c\u662f\u9488\u5bf9 Python 3.8 \u73af\u5883\u7684\u3002\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\u6765\u5b89\u88c5 PyTorch\uff1a<\/p>\n\n\n
python3<\/span>.8<\/span> -m<\/span> pip<\/span> install<\/span> torch<\/span> torchvision<\/span> torchaudio<\/span><\/code><\/span>Code language:<\/span> CSS<\/span> (<\/span>css<\/span>)<\/span><\/small><\/pre>\n\n\n
\u65b9\u6cd5\u4e8c\uff1a\u68c0\u67e5 <\/strong>pip<\/strong><\/code> \u7248\u672c\u5bf9\u5e94\u7684 Python \u89e3\u91ca\u5668<\/strong><\/h3>\n\n\n\n
\u4f60\u53ef\u4ee5\u5148\u67e5\u770b pip3<\/code> \u5bf9\u5e94\u7684 Python \u7248\u672c\uff1a<\/p>\n\n\n
pip3 -V<\/code><\/span><\/pre>\n\n\n
\u5982\u679c\u663e\u793a\u7684 Python \u7248\u672c\u662f 3.6\uff0c\u90a3\u4e48\u5c31\u8bf4\u660e\u5f53\u524d pip3<\/code> \u5173\u8054\u7684\u662f Python 3.6\u3002\u4f60\u9700\u8981\u4f7f\u7528 Python 3.8 \u5bf9\u5e94\u7684 pip<\/code> \u6765\u5b89\u88c5 PyTorch\u3002\u53ef\u4ee5\u901a\u8fc7\u4ee5\u4e0b\u65b9\u5f0f\u67e5\u770b Python 3.8 \u5bf9\u5e94\u7684 pip<\/code> \u8def\u5f84\uff1a<\/p>\n\n\n
which<\/span> python3<\/span>.8<\/span><\/code><\/span>Code language:<\/span> CSS<\/span> (<\/span>css<\/span>)<\/span><\/small><\/pre>\n\n\n
\u5047\u8bbe\u8f93\u51fa\u662f \/usr\/local\/bin\/python3.8<\/code>\uff0c\u90a3\u4e48\u5bf9\u5e94\u7684 pip<\/code> \u4e00\u822c\u662f \/usr\/local\/bin\/pip3.8<\/code>\uff08\u8def\u5f84\u53ef\u80fd\u56e0\u5b89\u88c5\u65b9\u5f0f\u4e0d\u540c\u800c\u6709\u6240\u5dee\u5f02\uff09\uff0c\u4f7f\u7528\u8be5 pip<\/code> \u6765\u5b89\u88c5 PyTorch\uff1a<\/p>\n\n\n
\/usr\/local\/bin\/pip3.8 install torch torchvision torchaudio<\/code><\/span><\/pre>\n\n\n
\u68c0\u67e5 GPU \u53ef\u7528\u6027<\/strong><\/h3>\n\n\n\n
\u5728\u89e3\u51b3\u4e0a\u8ff0\u95ee\u9898\u540e\uff0c\u4f60\u53ef\u4ee5\u518d\u6b21\u8fd0\u884c torch_check.py<\/code> \u811a\u672c\u6765\u68c0\u67e5 GPU \u662f\u5426\u53ef\u7528\u3002\u4ee5\u4e0b\u662f\u4e00\u4e2a\u7b80\u5355\u7684 torch_check.py<\/code> \u793a\u4f8b\u4ee3\u7801\uff1a<\/p>\n\n\n
import<\/span> torch<\/span>\n\nprint<\/span>(torch<\/span>.cuda<\/span>.is_available<\/span>())<\/code><\/span>Code language:<\/span> CSS<\/span> (<\/span>css<\/span>)<\/span><\/small><\/pre>\n\n\n
\u5982\u679c\u8f93\u51fa\u4e3a True<\/code>\uff0c\u5219\u8868\u793a GPU \u53ef\u4ee5\u6b63\u5e38\u4f7f\u7528\uff1b\u5982\u679c\u4ecd\u7136\u4e3a False<\/code>\uff0c\u5219\u9700\u8981\u8fdb\u4e00\u6b65\u6392\u67e5\u95ee\u9898\u3002<\/p>\n\n\n\n
\u5728\u4f7f\u7528 GPU \u8fdb\u884c\u8ba1\u7b97\u4efb\u52a1\uff08\u5982\u8fd0\u884c Whisper \u6a21\u578b\u8f6c\u5f55\u97f3\u9891\uff09\u65f6\uff0c\u5b9e\u65f6\u67e5\u770b GPU \u7684\u4f7f\u7528\u60c5\u51b5\u6709\u52a9\u4e8e\u4f60\u4e86\u89e3 GPU \u7684\u8d1f\u8f7d\u3001\u663e\u5b58\u5360\u7528\u7b49\u4fe1\u606f\uff0c\u4ee5\u4fbf\u4f18\u5316\u4efb\u52a1\u6216\u6392\u67e5\u95ee\u9898\u3002\u4ee5\u4e0b\u4ecb\u7ecd\u51e0\u79cd\u5e38\u89c1\u7684\u5b9e\u65f6\u67e5\u770b GPU \u4f7f\u7528\u60c5\u51b5\u7684\u65b9\u6cd5\uff1a<\/p>\n\n\n\n
1. \u4f7f\u7528 <\/strong>nvidia-smi<\/strong><\/code> \u547d\u4ee4\uff08\u9002\u7528\u4e8e NVIDIA GPU\uff09<\/strong><\/h3>\n\n\n\n
nvidia-smi<\/code> \u662f NVIDIA \u63d0\u4f9b\u7684\u4e00\u4e2a\u547d\u4ee4\u884c\u5de5\u5177\uff0c\u53ef\u7528\u4e8e\u76d1\u63a7 NVIDIA GPU \u7684\u72b6\u6001\u3002<\/p>\n\n\n\n
\u57fa\u672c\u4f7f\u7528<\/strong><\/h3>\n\n\n\n
\u6253\u5f00\u7ec8\u7aef\uff0c\u76f4\u63a5\u8f93\u5165 nvidia-smi<\/code> \u547d\u4ee4\uff0c\u5373\u53ef\u67e5\u770b\u5f53\u524d GPU \u7684\u57fa\u672c\u4fe1\u606f\uff0c\u5305\u62ec GPU \u7f16\u53f7\u3001\u540d\u79f0\u3001\u663e\u5b58\u4f7f\u7528\u60c5\u51b5\u3001GPU \u5229\u7528\u7387\u7b49\u3002<\/p>\n\n\n
nvidia-smi<\/code><\/span><\/pre>\n\n\n
\u793a\u4f8b\u8f93\u51fa\uff1a<\/p>\n\n\n
Thu Feb  7<\/span> 17<\/span>:00<\/span>:00<\/span> 2025<\/span>\n+-----------------------------------------------------------------------------+\n| NVIDIA-SMI 470.141<\/span>.03<\/span>   Driver Version: 470.141<\/span>.03<\/span>   CUDA Version: 11.4<\/span>     |\n|-------------------------------+----------------------+----------------------+\n| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n| Fan  Temp  Perf  Pwr:Usage\/Cap|         Memory-Usage | GPU-Util  Compute M. |\n|                               |                      |               MIG M. |\n|===============================+======================+======================|\n|   0<\/span>  Tesla P100-PCIE-16<\/span>GB    Off| 00000000<\/span>:00<\/span>:04.0<\/span> Off |                    0<\/span> |\n| N\/A   30<\/span>C    P0    26<\/span>W \/ 250<\/span>W |      5<\/span>MiB \/ 16384<\/span>MiB |      0<\/span>%      Default<\/span> |\n|                               |                      |                  N\/A |\n+-------------------------------+----------------------+----------------------+<\/code><\/span>Code language:<\/span> PHP<\/span> (<\/span>php<\/span>)<\/span><\/small><\/pre>\n\n\n
\u5b9e\u65f6\u76d1\u63a7<\/strong><\/h3>\n\n\n\n
\u82e5\u8981\u5b9e\u65f6\u67e5\u770b GPU \u4f7f\u7528\u60c5\u51b5\uff0c\u53ef\u4f7f\u7528 -l<\/code> \u6216 -lms<\/code> \u9009\u9879\u6307\u5b9a\u5237\u65b0\u95f4\u9694\u3002\u4f8b\u5982\uff0c\u6bcf\u9694 1 \u79d2\u5237\u65b0\u4e00\u6b21\uff1a<\/p>\n\n\n
nvidia-smi -l 1<\/code><\/span><\/pre>\n\n\n
\u6309 Ctrl + C<\/code> \u53ef\u505c\u6b62\u76d1\u63a7\u3002<\/p>\n\n\n\n
2. \u4f7f\u7528 <\/strong>gpustat<\/strong><\/code> \u5de5\u5177\uff08\u9002\u7528\u4e8e NVIDIA GPU\uff09<\/strong><\/h3>\n\n\n\n
gpustat<\/code> \u662f\u4e00\u4e2a\u57fa\u4e8e Python \u7684\u7b2c\u4e09\u65b9\u5de5\u5177\uff0c\u80fd\u4ee5\u66f4\u7f8e\u89c2\u548c\u6613\u8bfb\u7684\u65b9\u5f0f\u5c55\u793a GPU \u4fe1\u606f\u3002<\/p>\n\n\n\n
\u5b89\u88c5<\/strong><\/h3>\n\n\n\n
\u4f7f\u7528 pip<\/code> \u5b89\u88c5 gpustat<\/code>\uff1a<\/p>\n\n\n
python3<\/span>.8<\/span> -m<\/span> pip<\/span> install<\/span> gpustat<\/span><\/code><\/span>Code language:<\/span> CSS<\/span> (<\/span>css<\/span>)<\/span><\/small><\/pre>\n\n\n
\u4f7f\u7528<\/strong><\/h3>\n\n\n\n
\u5728\u7ec8\u7aef\u8f93\u5165 gpustat -i<\/code> \u547d\u4ee4\uff0c\u5373\u53ef\u5b9e\u65f6\u76d1\u63a7 GPU \u4f7f\u7528\u60c5\u51b5\uff0c\u9ed8\u8ba4\u5237\u65b0\u95f4\u9694\u4e3a 1 \u79d2\u3002<\/p>\n\n\n
gpustat -i<\/code><\/span><\/pre>\n\n\n
\u793a\u4f8b\u8f93\u51fa\uff1a<\/p>\n\n\n
[0] Tesla P100-PCIE-16GB | 30\u00b0C,   0 % |    5 \/ 16384 MB |<\/code><\/span><\/pre>\n\n\n
\u793a\u4f8b\u4ee3\u7801\uff1a<\/p>\n\n\n
import os\nimport subprocess\nimport time\nimport logging\nimport requests\nimport urllib.parse\nimport datetime\n\n# \u914d\u7f6e\u6839\u6587\u4ef6\u5939<\/span>\nroot_folder = \"\/root\/audio_files\"<\/span>\n\n# \u914d\u7f6e\u65e5\u5fd7\u6587\u4ef6\u8def\u5f84<\/span>\nlog_file = os.path.join(root_folder, \"transcription.log\"<\/span>)\n\n# \u914d\u7f6e\u65e5\u5fd7\u8bb0\u5f55<\/span>\nlogger = logging.getLogger(__name__)\nlogger.setLevel(logging.INFO)\n\n# \u521b\u5efa\u6587\u4ef6\u5904\u7406\u5668<\/span>\nfile_handler = logging.FileHandler(log_file)\nfile_handler.setLevel(logging.INFO)\n\n# \u521b\u5efa\u63a7\u5236\u53f0\u5904\u7406\u5668<\/span>\nconsole_handler = logging.StreamHandler()\nconsole_handler.setLevel(logging.INFO)\n\n# \u5b9a\u4e49\u65e5\u5fd7\u683c\u5f0f<\/span>\nformatter = logging.Formatter('%(asctime)s - %(levelname)s - %(message)s'<\/span>)\nfile_handler.setFormatter(formatter)\nconsole_handler.setFormatter(formatter)\n\n# \u5c06\u5904\u7406\u5668\u6dfb\u52a0\u5230\u65e5\u5fd7\u8bb0\u5f55\u5668<\/span>\nlogger.addHandler(file_handler)\nlogger.addHandler(console_handler)\n\n# \u8bed\u8a00\u3001\u6a21\u578b\u548c\u7ebf\u7a0b\u6570\u914d\u7f6e<\/span>\nlanguage = \"Japanese\"<\/span>\nmodel_name = \"medium\"<\/span>\nthreads = 4<\/span>  # \u8bbe\u7f6e\u7ebf\u7a0b\u6570<\/span>\n\n# \u652f\u6301\u7684\u97f3\u9891\u6587\u4ef6\u6269\u5c55\u540d<\/span>\naudio_extensions = (\".mp3\"<\/span>, \".wav\"<\/span>, \".ogg\"<\/span>, \".flac\"<\/span>)\n\n# \u8f6c\u5f55\u97f3\u9891\u6587\u4ef6\u5e76\u4fdd\u5b58\u7ed3\u679c<\/span>\ndef transcribe_audio(audio_file, sub_folder_path):\n    # \u83b7\u53d6\u6587\u4ef6\u540d\uff08\u4e0d\u5305\u542b\u6269\u5c55\u540d\uff09<\/span>\n    file_name = os.path.splitext(os.path.basename(audio_file))[0<\/span>]\n    # \u751f\u6210\u6bcf\u4e2a\u5b50\u6587\u4ef6\u5939\u4e0b\u7684\u8f93\u51fa\u6587\u4ef6\u5939\u8def\u5f84<\/span>\n    output_folder = os.path.join(sub_folder_path, \"TranscriptionResults\"<\/span>)\n    # \u68c0\u67e5\u8f93\u51fa\u6587\u4ef6\u5939\u662f\u5426\u5b58\u5728\uff0c\u4e0d\u5b58\u5728\u5219\u521b\u5efa<\/span>\n    if<\/span> not os.path.isdir(output_folder):\n        os.makedirs(output_folder)\n    # \u751f\u6210\u8f93\u51fa\u6587\u4ef6\u7684\u57fa\u672c\u8def\u5f84<\/span>\n    output_base_path = os.path.join(output_folder, file_name)\n\n    # \u8bb0\u5f55\u5f00\u59cb\u65f6\u95f4<\/span>\n    start_time = time.time()\n    logger.info(f\"\u6b63\u5728\u8f6c\u5f55\u6587\u4ef6: {audio_file}\"<\/span>)\n    try<\/span>:\n        # \u901a\u8fc7 subprocess \u8c03\u7528 Whisper \u547d\u4ee4<\/span>\n        command = [\n            \"whisper\"<\/span>, audio_file,\n            \"--language\"<\/span>, language,\n            \"--model\"<\/span>, model_name,\n            \"--threads\"<\/span>, str(threads),\n            \"--output_dir\"<\/span>, output_folder  # \u6307\u5b9a\u8f93\u51fa\u76ee\u5f55<\/span>\n        ]\n\n        # \u6267\u884c\u547d\u4ee4<\/span>\n        subprocess.run(command, check=True<\/span>)\n\n        # \u8bb0\u5f55\u7ed3\u675f\u65f6\u95f4\u5e76\u8ba1\u7b97\u8f6c\u5f55\u65f6\u95f4<\/span>\n        end_time = time.time()\n        elapsed_time = end_time - start_time\n        logger.info(f\"\u8f6c\u5f55\u5b8c\u6210\uff0c\u6240\u6709\u7ed3\u679c\u5df2\u4fdd\u5b58\u5230: {output_folder}\"<\/span>)\n        logger.info(f\"\u6587\u4ef6 {audio_file} \u7684\u8f6c\u5f55\u82b1\u8d39\u4e86 {elapsed_time:.2f} \u79d2\"<\/span>)\n    except subprocess.CalledProcessError as<\/span> e:\n        logger.error(f\"\u8f6c\u5f55\u6587\u4ef6 {audio_file} \u65f6\u51fa\u9519: {e}\"<\/span>)\n    finally<\/span>:\n        pass\n\n# \u683c\u5f0f\u5316\u603b\u65f6\u95f4\u4e3a \"\u5929:\u5c0f\u65f6:\u5206\u949f:\u79d2\" \u683c\u5f0f<\/span>\ndef format_total_time(seconds):\n    days = seconds \/\/ (24 * 3600)<\/span>\n    hours = (seconds % (24<\/span> * 3600<\/span>)) \/\/ 3600<\/span>\n    minutes = (seconds % 3600<\/span>) \/\/ 60<\/span>\n    seconds = seconds % 60<\/span>\n    return<\/span> f\"{int(days)}\u5929 {int(hours)}\u5c0f\u65f6 {int(minutes)}\u5206\u949f {int(seconds)}\u79d2\"<\/span>\n\n# \u63a8\u9001\u6d88\u606f\u5230PushDeer<\/span>\ndef push_to_pushdeer(message, api_key, push_deer_url, push_title, push_type):\n    timestamp = datetime.datetime.now().strftime(\"%Y-%m-%d %H:%M:%S\"<\/span>)\n    try<\/span>:\n        encoded_message = urllib.parse.quote(message, safe=''<\/span>)\n        request_body = {\n            \"pushkey\"<\/span>: api_key,\n            \"text\"<\/span>: f\"{push_title} {timestamp}\"<\/span>,\n            \"desp\"<\/span>: encoded_message,\n            \"type\"<\/span>: push_type\n        }\n        response = requests.post(push_deer_url, data=request_body)\n\n        if<\/span> response.status_code == 200<\/span>:\n            logger.info(\"\u63a8\u9001\u6210\u529f,\u65f6\u95f4\u4e3a\uff1a{}\"<\/span>.format(timestamp))\n        else<\/span>:\n            logger.error(f\"\u63a8\u9001\u5931\u8d25\uff0c\u72b6\u6001\u7801: {response.status_code}\"<\/span>)\n    except Exception<\/span> as<\/span> e:\n        logger.error(f\"\u63a8\u9001\u5931\u8d25\uff1a{e}\"<\/span>)\n\n# \u4e3b\u7a0b\u5e8f<\/span>\nif<\/span> __name__ == \"__main__\"<\/span>:\n    # \u6536\u96c6\u6240\u6709\u97f3\u9891\u6587\u4ef6\u8def\u5f84\u53ca\u5176\u6240\u5728\u5b50\u6587\u4ef6\u5939\u8def\u5f84<\/span>\n    audio_file_info = []\n    for<\/span> sub_folder in os.listdir(root_folder):\n        sub_folder_path = os.path.join(root_folder, sub_folder)\n        if<\/span> os.path.isdir(sub_folder_path):\n            for<\/span> root_dir, _, files in os.walk(sub_folder_path):\n                for<\/span> file in files:\n                    if<\/span> file.lower().endswith(audio_extensions):\n                        audio_file = os.path.join(root_dir, file)\n                        audio_file_info.append((audio_file, sub_folder_path))\n\n    # \u8bb0\u5f55\u603b\u5f00\u59cb\u65f6\u95f4<\/span>\n    total_start_time = time.time()\n\n    # \u4f7f\u7528\u7ebf\u7a0b\u6c60\u5e76\u53d1\u5904\u7406\u97f3\u9891\u8f6c\u5f55<\/span>\n    for<\/span> audio_file, sub_folder_path in audio_file_info:\n        transcribe_audio(audio_file, sub_folder_path)\n\n    # \u8bb0\u5f55\u603b\u7ed3\u675f\u65f6\u95f4<\/span>\n    total_end_time = time.time()\n    total_elapsed_time = total_end_time - total_start_time\n\n    # \u8f93\u51fa\u603b\u8f6c\u5f55\u65f6\u95f4<\/span>\n    logger.info(f\"\\n\u6240\u6709\u97f3\u9891\u6587\u4ef6\u7684\u8f6c\u5f55\u5168\u90e8\u5b8c\u6210\uff0c\u82b1\u8d39\u4e86 {format_total_time(total_elapsed_time)}\"<\/span>)\n\n    # \u63a8\u9001\u6d88\u606f<\/span>\n    message = f\"\u6240\u6709\u97f3\u9891\u6587\u4ef6\u8f6c\u5f55\u5df2\u5b8c\u6210\uff0c\u5171\u82b1\u8d39\u4e86 {format_total_time(total_elapsed_time)}\"<\/span>\n    api_key = \"PDU1TVFy1tBulCz7Hq1ajeFqe6ccPwSbgw09x\"<\/span>\n    push_deer_url = \"http:\/\/43.153.151.226:8800\/message\/push\"<\/span>\n    push_title = \"\u8f6c\u5f55\u52a8\u6001\u63a8\u9001\"<\/span>\n    push_type = \"markdown\"<\/span>\n\n    push_to_pushdeer(message, api_key, push_deer_url, push_title, push_type)<\/code><\/span>Code language:<\/span> PHP<\/span> (<\/span>php<\/span>)<\/span><\/small><\/pre>","protected":false},"excerpt":{"rendered":"
Whisper \u662f OpenAI \u5f00\u53d1\u7684\u4e00\u4e2a\u81ea\u52a8\u8bed\u97f3\u8bc6\u522b\uff08ASR\uff09\u7cfb\u7edf\uff0c\u5b83\u63d0\u4f9b\u4e86\u4e0d\u540c\u5927\u5c0f\u7684\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u6a21\u578b\u8d8a\u5927 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_aikuto_target_date":"","footnotes":""},"categories":[2],"tags":[33,30,32],"class_list":["post-1155","post","type-post","status-publish","format-standard","hentry","category-sketches","tag-asr","tag-whisper","tag-32"],"_links":{"self":[{"href":"https:\/\/aikuto.com\/index.php?rest_route=\/wp\/v2\/posts\/1155","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aikuto.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aikuto.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aikuto.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/aikuto.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1155"}],"version-history":[{"count":2,"href":"https:\/\/aikuto.com\/index.php?rest_route=\/wp\/v2\/posts\/1155\/revisions"}],"predecessor-version":[{"id":1159,"href":"https:\/\/aikuto.com\/index.php?rest_route=\/wp\/v2\/posts\/1155\/revisions\/1159"}],"wp:attachment":[{"href":"https:\/\/aikuto.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1155"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aikuto.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1155"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aikuto.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1155"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}