{ "version": "https://jsonfeed.org/version/1.1", "user_comment": "This feed allows you to read the posts from this site in any feed reader that supports the JSON Feed format. To add this feed to your reader, copy the following URL -- https://matoken.org/blog/tag/whisper/feed/json/ -- and add it your reader.", "home_page_url": "https://matoken.org/blog/tag/whisper/", "feed_url": "https://matoken.org/blog/tag/whisper/feed/json/", "language": "ja", "title": "whisper – matoken's blog", "description": "Is there no plan B?", "icon": "https://matoken.org/blog/wp-content/uploads/2025/03/cropped-1865f695c4eecc844385acef2f078255036adccd42c254580ea3844543ab56d9.jpeg", "items": [ { "id": "https://matoken.org/blog/?p=4125", "url": "https://matoken.org/blog/2024/10/11/trying-out-openai-whispers-new-large-v3-turbo-model/", "title": "OpenAI Whisper\u306e\u65b0\u3057\u3044large-v3-turbo\u30e2\u30c7\u30eb\u3092\u5c11\u3057\u8a66\u3059", "content_html": "

\n
\n

OpenAI \u306e Whisper \u3068\u3044\u3046\u6587\u5b57\u8d77\u3053\u3057\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u304c\u3042\u308a\u307e\u3059\uff0e\u3053\u308c\u3092\u4f7f\u3044\uff0c\u30ab\u30f3\u30d5\u30a1\u30ec\u30f3\u30b9\u7b49\u306e\u6587\u5b57\u8d77\u3053\u3057\u3092\u3057\u305f\u308a\u3057\u3066\u4fbf\u5229\u306b\u4f7f\u308f\u305b\u3066\u3082\u3089\u3063\u3066\u3044\u307e\u3059\uff0e

\n
\n
\n\n
\n
\n

Whisper \u3092\u4f7f\u3046\u969b\u306b\u30e2\u30c7\u30eb\u3092\u6307\u5b9a\u3059\u308b\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\uff0e
\n\u30e2\u30c7\u30eb\u306f\u8907\u6570\u5b58\u5728\u3057\uff0c\u57fa\u672c\u7684\u306b\u30b5\u30a4\u30ba\u306e\u5927\u304d\u3044\u3082\u306e\u306e\u307b\u3046\u304c\u6587\u5b57\u8d77\u3053\u3057\u7cbe\u5ea6\u304c\u826f\u3044\u3067\u3059\u304c\uff0c\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u306e\u30ea\u30bd\u30fc\u30b9\u3092\u6d88\u8cbb\u3057\u307e\u3059\uff0e
\n\u4eca\u56de large-v3-turbo(turbo \u3068\u7565\u3055\u308c\u308b\u3088\u3046\u3067\u3059\uff0e)\u3068\u3044\u3046\u65b0\u3057\u3044\u30e2\u30c7\u30eb\u304c\u30ea\u30ea\u30fc\u30b9\u3055\u308c\u307e\u3057\u305f\uff0e\u3053\u306e\u30e2\u30c7\u30eb\u306f large-v3 \u3092\u5143\u306b\u30c7\u30b3\u30fc\u30c0\u3092\u5c0f\u3055\u304f\u3059\u308b\u3053\u3068\u3067\u7cbe\u5ea6\u306f\u305d\u306e\u307e\u307e\u306b\u9ad8\u901f\u306b\u306a\u3063\u3066\u3044\u308b\u305d\u3046\u3067\u3059\uff0e
\n\u3069\u306e\u304f\u3089\u3044\u901f\u304f\u306a\u3063\u3066\u3044\u308b\u306e\u304b\u5c11\u3057\u8a66\u3057\u3066\u307f\u307e\u3057\u305f\uff0e

\n
\n
\n\n
\n

\n
\n

\u4ee5\u4e0b\u306f Whisper \u306e README.md \u306b\u3042\u308b\u30e2\u30c7\u30eb\u306e\u6bd4\u8f03\u3067\u3059\uff0e\u4e00\u756a\u4e0b\u306b\u8ffd\u52a0\u3055\u308c\u305f turbo \u304c\u4eca\u56de\u8a66\u3057\u305f\u3044\u3082\u306e\u3067\u3059\uff0e

\n
\n
\n
\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
SizeParametersEnglish-only modelMultilingual modelRequired VRAMRelative speed
tiny39 Mtiny.entiny~1 GB~10x
base74 Mbase.enbase~1 GB~7x
small244 Msmall.ensmall~2 GB~4x
medium769 Mmedium.enmedium~5 GB~2x
large1550 MN/Alarge~10 GB1x
turbo809 MN/Aturbo~6 GB~8x
\n
\n
\n— Available models and languages\n
\n
\n
\n

turbo \u3092\u8a66\u3059\u306b\u306f Python \u306e openai-whisper \u30d1\u30c3\u30b1\u30fc\u30b8\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092 20240930 \u4ee5\u964d\u306b\u3059\u308b\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\uff0e
\n\u65b0\u898f\u306b\u5c0e\u5165\u3059\u308b\u5834\u5408\u306f\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u611f\u3058\u3067\uff0e

\n
\n
\n
\n
$ python3 -m venv venv\n$ source venv/bin/activate\n$ pip install openai-whisper
\n
\n
\n
\n

\u65e2\u5b58\u306e\u3082\u306e\u3092\u30a2\u30c3\u30d7\u30b0\u30ec\u30fc\u30c9\u3059\u308b\u5834\u5408\u306f\u3053\u3093\u306a\u611f\u3058\u3067\uff0e

\n
\n
\n
\n
$ source venv/bin/activate\n$ pip install -U openai-whisper
\n
\n
\n
\n

\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u78ba\u8a8d\u3057\u3066 20240930 \u4ee5\u964d\u306b\u306a\u3063\u3066\u3044\u308b\u3053\u3068\u3092\u78ba\u8a8d\uff0e

\n
\n
\n
\n
$ pip list | grep whisper\nopenai-whisper           20240930
\n
\n
\n
\n

\u30c6\u30b9\u30c8\u306e\u305f\u3081\u306e\u97f3\u58f0\u3092\u7528\u610f\u3057\u307e\u3059\uff0e\u4eca\u56de\u306fNHK\u30e9\u30b8\u30aa\u30cb\u30e5\u30fc\u30b9\u3092\u5229\u7528\u3057\u307e\u3057\u305f\uff0eRSS feed \u304b\u3089\u6700\u65b0\u306e\u30cb\u30e5\u30fc\u30b9\u3092\u5165\u624b\u3057\u307e\u3057\u305f\uff0e

\n
\n
\n
\n
$ rsstail -n1 -N1eu https://www.nhk.or.jp/s-media/news/podcast/list/v1/all.xml | grep .mp3$ | xargs wget
\n
\n
\n
\n

\u97f3\u58f0\u60c5\u5831\uff0e\u4e01\u5ea65\u5206\u307b\u3069\u306e\u97f3\u58f0\u3067\u3059\uff0e

\n
\n
\n
\n
$ ffprobe 08fe3ee2f8d9f9fa8481508530974854_64k.mp3 2>&1 |grep Stream\n  Stream #0:0: Audio: mp3 (mp3float), 48000 Hz, stereo, fltp, 64 kb/s\n$ ffprobe 08fe3ee2f8d9f9fa8481508530974854_64k.mp3 2>&1 |grep Duration:\n  Duration: 00:05:00.02, start: 0.023021, bitrate: 64 kb/s
\n
\n
\n
\n

\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u304c\u306a\u3044\u72b6\u614b\u3067 Whisper \u3092\u5b9f\u884c\u3059\u308b\u3068\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u304c\u5b9f\u884c\u3055\u308c\u307e\u3059\uff0e

\n
\n
\n

Linux \u74b0\u5883\u3060\u3068\u65e2\u5b9a\u5024\u3067\u306f ~/.cache/whisper/ \u306b\u4fdd\u5b58\u3055\u308c\u307e\u3059\uff0e--model_dir \u30aa\u30d7\u30b7\u30e7\u30f3\u3067\u5909\u66f4\u3059\u308b\u3053\u3068\u3082\u53ef\u80fd\u3067\u3059\uff0e

\n
\n
\n

\u30b5\u30a4\u30ba\u3092\u898b\u308b\u3068 turbo \u306f 1.6GB \u7a0b\u3067 medium \u3068\u30b5\u30a4\u30ba\u304c\u8fd1\u3044\u3067\u3059\uff0e

\n
\n
\n
\n
$ ls -Ss1 ~/.cache/whisper/\ntotal 6990496\n3014656 large.pt\n1580028 large-v3-turbo.pt\n1492200 medium.pt\n 472284 small.pt\n 141860 base.pt\n 141860 base.en.pt\n  73804 tiny.pt\n  73804 tiny.en.pt
\n
\n
\n
\n

turbo \u3068 medium \u3067\u5b9f\u884c\u3057\u3066\u307f\u307e\u3057\u305f\uff0e
\n\u4eca\u56de\u5b9f\u884c\u3057\u305f\u7aef\u672b\u306f CPU:Intel Core i7-10510U/RAM 16GB \u306e\u7aef\u672b\u3067\u3059\uff0e\u65e5\u5e38\u4f5c\u696d\u3092\u884c\u3044\u306a\u304c\u3089\u3060\u3063\u305f\u306e\u3067 Whisper \u3060\u3051\u3067\u52d5\u304b\u3059\u3068\u3082\u3046\u5c11\u3057\u901f\u304f\u306a\u308b\u3068\u601d\u3044\u307e\u3059\uff0e

\n
\n
\n
turbo
\n
\n
$ time whisper --model large-v3-turbo --threads `nproc` --language Japanese 08fe3ee2f8d9f9fa8481508530974854_64k.mp3\n\u301c\u7565\u301c\n\nreal    6m48.087s\nuser    46m41.336s\nsys     2m31.058s
\n
\n
\n
\n
medium
\n
\n
$ time whisper --model medium --threads `nproc` --language Japanese 08fe3ee2f8d9f9fa8481508530974854_64k.mp3\n\u301c\u7565\u301c\n\nreal    19m49.721s\nuser    145m13.502s\nsys     0m44.894s
\n
\n
\n
\n

turbo \u306f\u5927\u5206\u901f\u3044\u3067\u3059\u306d\uff0e\u3082\u3046\u5c11\u3057\u65e9\u304f\u306a\u308b\u3068\u30ea\u30a2\u30eb\u30bf\u30a4\u30e0\u6587\u5b57\u8d77\u3053\u3057\u304c\u3067\u304d\u308b\u306e\u3067\u3059\u304c\uff0e
\n\u7d50\u679c\u306e\u6587\u5b57\u8d77\u3053\u3057\u306e\u7cbe\u5ea6\u3082\u554f\u984c\u306a\u3055\u305d\u3046\u3067\u3059\uff0e\u3057\u3070\u3089\u304f\u306f\u3053\u306e large-v3-turbo \u30e2\u30c7\u30eb\u3092\u4f7f\u3063\u3066\u3044\u3053\u3046\u3068\u601d\u3044\u307e\u3059\uff0e

\n
\n
\n
\u74b0\u5883
\n
\n
$ pip list | grep whisper\nopenai-whisper           20240930\n$ dpkg-query -W python3 rsstail\npython3 3.12.6-1\nrsstail 2.1-1\n$ lsb_release -dr\nDescription:    Debian GNU/Linux trixie/sid\nRelease:        n/a\n$ arch\nx86_64\n$ cpu-info\nPackages:\n        0: Intel Core i7-10510U\nMicroarchitectures:\n        4x Sky Lake\nCores:\n        0: 2 processors (0-1), Intel Sky Lake\n        1: 2 processors (2-3), Intel Sky Lake\n        2: 2 processors (4-5), Intel Sky Lake\n        3: 2 processors (6-7), Intel Sky Lake\nClusters:\n        0: 8 processors (0-7),  0: 4 cores (0-3), Intel Sky Lake\nLogical processors (System ID):\n        0 (0): APIC ID 0x00000000\n        1 (4): APIC ID 0x00000001\n        2 (1): APIC ID 0x00000002\n        3 (5): APIC ID 0x00000003\n        4 (2): APIC ID 0x00000004\n        5 (6): APIC ID 0x00000005\n        6 (3): APIC ID 0x00000006\n        7 (7): APIC ID 0x00000007\n$ grep ^MemTotal: /proc/meminfo\nMemTotal:       15987016 kB\n$ free\n               total        used        free      shared  buff/cache   available\nMem:        15987016     9784984     2546356     2600260     6612472     6202032\nSwap:       17575932      627800    16948132
\n
\n
\n", "content_text": "OpenAI \u306e Whisper \u3068\u3044\u3046\u6587\u5b57\u8d77\u3053\u3057\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u304c\u3042\u308a\u307e\u3059\uff0e\u3053\u308c\u3092\u4f7f\u3044\uff0c\u30ab\u30f3\u30d5\u30a1\u30ec\u30f3\u30b9\u7b49\u306e\u6587\u5b57\u8d77\u3053\u3057\u3092\u3057\u305f\u308a\u3057\u3066\u4fbf\u5229\u306b\u4f7f\u308f\u305b\u3066\u3082\u3089\u3063\u3066\u3044\u307e\u3059\uff0e\n\n\n\n\nopenai/whisper: Robust Speech Recognition via Large-Scale Weak Supervision\n\n\nOpenAI\u306eWhisper\u3067\u30aa\u30d5\u30e9\u30a4\u30f3\u6587\u5b57\u8d77\u3053\u3057(STT) (\u9e7f\u5150\u5cf6Linux\u52c9\u5f37\u4f1a 2022.11\u306b\u3066\u767a\u8868)\n\n\n\n\nWhisper \u3092\u4f7f\u3046\u969b\u306b\u30e2\u30c7\u30eb\u3092\u6307\u5b9a\u3059\u308b\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\uff0e\n\u30e2\u30c7\u30eb\u306f\u8907\u6570\u5b58\u5728\u3057\uff0c\u57fa\u672c\u7684\u306b\u30b5\u30a4\u30ba\u306e\u5927\u304d\u3044\u3082\u306e\u306e\u307b\u3046\u304c\u6587\u5b57\u8d77\u3053\u3057\u7cbe\u5ea6\u304c\u826f\u3044\u3067\u3059\u304c\uff0c\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u306e\u30ea\u30bd\u30fc\u30b9\u3092\u6d88\u8cbb\u3057\u307e\u3059\uff0e\n\u4eca\u56de large-v3-turbo(turbo \u3068\u7565\u3055\u308c\u308b\u3088\u3046\u3067\u3059\uff0e)\u3068\u3044\u3046\u65b0\u3057\u3044\u30e2\u30c7\u30eb\u304c\u30ea\u30ea\u30fc\u30b9\u3055\u308c\u307e\u3057\u305f\uff0e\u3053\u306e\u30e2\u30c7\u30eb\u306f large-v3 \u3092\u5143\u306b\u30c7\u30b3\u30fc\u30c0\u3092\u5c0f\u3055\u304f\u3059\u308b\u3053\u3068\u3067\u7cbe\u5ea6\u306f\u305d\u306e\u307e\u307e\u306b\u9ad8\u901f\u306b\u306a\u3063\u3066\u3044\u308b\u305d\u3046\u3067\u3059\uff0e\n\u3069\u306e\u304f\u3089\u3044\u901f\u304f\u306a\u3063\u3066\u3044\u308b\u306e\u304b\u5c11\u3057\u8a66\u3057\u3066\u307f\u307e\u3057\u305f\uff0e\n\n\n\n\nturbo model release \u00b7 openai/whisper \u00b7 Discussion #2363\n\n\n\n\n\n\u4ee5\u4e0b\u306f Whisper \u306e README.md \u306b\u3042\u308b\u30e2\u30c7\u30eb\u306e\u6bd4\u8f03\u3067\u3059\uff0e\u4e00\u756a\u4e0b\u306b\u8ffd\u52a0\u3055\u308c\u305f turbo \u304c\u4eca\u56de\u8a66\u3057\u305f\u3044\u3082\u306e\u3067\u3059\uff0e\n\n\n\n\n\n\n\n\n\n\n\n\n\n\nSize\nParameters\nEnglish-only model\nMultilingual model\nRequired VRAM\nRelative speed\n\n\n\n\ntiny\n39 M\ntiny.en\ntiny\n~1 GB\n~10x\n\n\nbase\n74 M\nbase.en\nbase\n~1 GB\n~7x\n\n\nsmall\n244 M\nsmall.en\nsmall\n~2 GB\n~4x\n\n\nmedium\n769 M\nmedium.en\nmedium\n~5 GB\n~2x\n\n\nlarge\n1550 M\nN/A\nlarge\n~10 GB\n1x\n\n\nturbo\n809 M\nN/A\nturbo\n~6 GB\n~8x\n\n\n\n\n\n— Available models and languages\n\n\n\nturbo \u3092\u8a66\u3059\u306b\u306f Python \u306e openai-whisper \u30d1\u30c3\u30b1\u30fc\u30b8\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092 20240930 \u4ee5\u964d\u306b\u3059\u308b\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\uff0e\n\u65b0\u898f\u306b\u5c0e\u5165\u3059\u308b\u5834\u5408\u306f\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u611f\u3058\u3067\uff0e\n\n\n\n$ python3 -m venv venv\n$ source venv/bin/activate\n$ pip install openai-whisper\n\n\n\n\u65e2\u5b58\u306e\u3082\u306e\u3092\u30a2\u30c3\u30d7\u30b0\u30ec\u30fc\u30c9\u3059\u308b\u5834\u5408\u306f\u3053\u3093\u306a\u611f\u3058\u3067\uff0e\n\n\n\n$ source venv/bin/activate\n$ pip install -U openai-whisper\n\n\n\n\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u78ba\u8a8d\u3057\u3066 20240930 \u4ee5\u964d\u306b\u306a\u3063\u3066\u3044\u308b\u3053\u3068\u3092\u78ba\u8a8d\uff0e\n\n\n\n$ pip list | grep whisper\nopenai-whisper 20240930\n\n\n\n\u30c6\u30b9\u30c8\u306e\u305f\u3081\u306e\u97f3\u58f0\u3092\u7528\u610f\u3057\u307e\u3059\uff0e\u4eca\u56de\u306fNHK\u30e9\u30b8\u30aa\u30cb\u30e5\u30fc\u30b9\u3092\u5229\u7528\u3057\u307e\u3057\u305f\uff0eRSS feed \u304b\u3089\u6700\u65b0\u306e\u30cb\u30e5\u30fc\u30b9\u3092\u5165\u624b\u3057\u307e\u3057\u305f\uff0e\n\n\n\n$ rsstail -n1 -N1eu https://www.nhk.or.jp/s-media/news/podcast/list/v1/all.xml | grep .mp3$ | xargs wget\n\n\n\n\u97f3\u58f0\u60c5\u5831\uff0e\u4e01\u5ea65\u5206\u307b\u3069\u306e\u97f3\u58f0\u3067\u3059\uff0e\n\n\n\n$ ffprobe 08fe3ee2f8d9f9fa8481508530974854_64k.mp3 2>&1 |grep Stream\n Stream #0:0: Audio: mp3 (mp3float), 48000 Hz, stereo, fltp, 64 kb/s\n$ ffprobe 08fe3ee2f8d9f9fa8481508530974854_64k.mp3 2>&1 |grep Duration:\n Duration: 00:05:00.02, start: 0.023021, bitrate: 64 kb/s\n\n\n\n\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u304c\u306a\u3044\u72b6\u614b\u3067 Whisper \u3092\u5b9f\u884c\u3059\u308b\u3068\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u304c\u5b9f\u884c\u3055\u308c\u307e\u3059\uff0e\n\n\nLinux \u74b0\u5883\u3060\u3068\u65e2\u5b9a\u5024\u3067\u306f ~/.cache/whisper/ \u306b\u4fdd\u5b58\u3055\u308c\u307e\u3059\uff0e--model_dir \u30aa\u30d7\u30b7\u30e7\u30f3\u3067\u5909\u66f4\u3059\u308b\u3053\u3068\u3082\u53ef\u80fd\u3067\u3059\uff0e\n\n\n\u30b5\u30a4\u30ba\u3092\u898b\u308b\u3068 turbo \u306f 1.6GB \u7a0b\u3067 medium \u3068\u30b5\u30a4\u30ba\u304c\u8fd1\u3044\u3067\u3059\uff0e\n\n\n\n$ ls -Ss1 ~/.cache/whisper/\ntotal 6990496\n3014656 large.pt\n1580028 large-v3-turbo.pt\n1492200 medium.pt\n 472284 small.pt\n 141860 base.pt\n 141860 base.en.pt\n 73804 tiny.pt\n 73804 tiny.en.pt\n\n\n\nturbo \u3068 medium \u3067\u5b9f\u884c\u3057\u3066\u307f\u307e\u3057\u305f\uff0e\n\u4eca\u56de\u5b9f\u884c\u3057\u305f\u7aef\u672b\u306f CPU:Intel Core i7-10510U/RAM 16GB \u306e\u7aef\u672b\u3067\u3059\uff0e\u65e5\u5e38\u4f5c\u696d\u3092\u884c\u3044\u306a\u304c\u3089\u3060\u3063\u305f\u306e\u3067 Whisper \u3060\u3051\u3067\u52d5\u304b\u3059\u3068\u3082\u3046\u5c11\u3057\u901f\u304f\u306a\u308b\u3068\u601d\u3044\u307e\u3059\uff0e\n\n\nturbo\n\n$ time whisper --model large-v3-turbo --threads `nproc` --language Japanese 08fe3ee2f8d9f9fa8481508530974854_64k.mp3\n\u301c\u7565\u301c\n\nreal 6m48.087s\nuser 46m41.336s\nsys 2m31.058s\n\n\n\nmedium\n\n$ time whisper --model medium --threads `nproc` --language Japanese 08fe3ee2f8d9f9fa8481508530974854_64k.mp3\n\u301c\u7565\u301c\n\nreal 19m49.721s\nuser 145m13.502s\nsys 0m44.894s\n\n\n\nturbo \u306f\u5927\u5206\u901f\u3044\u3067\u3059\u306d\uff0e\u3082\u3046\u5c11\u3057\u65e9\u304f\u306a\u308b\u3068\u30ea\u30a2\u30eb\u30bf\u30a4\u30e0\u6587\u5b57\u8d77\u3053\u3057\u304c\u3067\u304d\u308b\u306e\u3067\u3059\u304c\uff0e\n\u7d50\u679c\u306e\u6587\u5b57\u8d77\u3053\u3057\u306e\u7cbe\u5ea6\u3082\u554f\u984c\u306a\u3055\u305d\u3046\u3067\u3059\uff0e\u3057\u3070\u3089\u304f\u306f\u3053\u306e large-v3-turbo \u30e2\u30c7\u30eb\u3092\u4f7f\u3063\u3066\u3044\u3053\u3046\u3068\u601d\u3044\u307e\u3059\uff0e\n\n\n\u74b0\u5883\n\n$ pip list | grep whisper\nopenai-whisper 20240930\n$ dpkg-query -W python3 rsstail\npython3 3.12.6-1\nrsstail 2.1-1\n$ lsb_release -dr\nDescription: Debian GNU/Linux trixie/sid\nRelease: n/a\n$ arch\nx86_64\n$ cpu-info\nPackages:\n 0: Intel Core i7-10510U\nMicroarchitectures:\n 4x Sky Lake\nCores:\n 0: 2 processors (0-1), Intel Sky Lake\n 1: 2 processors (2-3), Intel Sky Lake\n 2: 2 processors (4-5), Intel Sky Lake\n 3: 2 processors (6-7), Intel Sky Lake\nClusters:\n 0: 8 processors (0-7), 0: 4 cores (0-3), Intel Sky Lake\nLogical processors (System ID):\n 0 (0): APIC ID 0x00000000\n 1 (4): APIC ID 0x00000001\n 2 (1): APIC ID 0x00000002\n 3 (5): APIC ID 0x00000003\n 4 (2): APIC ID 0x00000004\n 5 (6): APIC ID 0x00000005\n 6 (3): APIC ID 0x00000006\n 7 (7): APIC ID 0x00000007\n$ grep ^MemTotal: /proc/meminfo\nMemTotal: 15987016 kB\n$ free\n total used free shared buff/cache available\nMem: 15987016 9784984 2546356 2600260 6612472 6202032\nSwap: 17575932 627800 16948132", "date_published": "2024-10-11T06:29:54+09:00", "date_modified": "2024-10-11T06:29:55+09:00", "authors": [ { "name": "matoken", "url": "https://matoken.org/blog/author/matoken/", "avatar": "https://secure.gravatar.com/avatar/38f5f3b575c5eb45cda6aa659bca119ac7a5e16b46565e869d0030e3bd66981d?s=512&d=mm&r=g" } ], "author": { "name": "matoken", "url": "https://matoken.org/blog/author/matoken/", "avatar": "https://secure.gravatar.com/avatar/38f5f3b575c5eb45cda6aa659bca119ac7a5e16b46565e869d0030e3bd66981d?s=512&d=mm&r=g" }, "tags": [ "transcription", "whisper", "Debian", "Linux", "sid" ] }, { "id": "https://matoken.org/blog/?p=4062", "url": "https://matoken.org/blog/2024/08/02/downloading-the-model-data-of-whisper-cpp-fails-and-then-re-downloading-it/", "title": "whisper.cpp \u306e\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u306b\u5931\u6557\u3057\u305f\u3042\u3068\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u76f4\u3059", "content_html": "
\n

OpenAI \u306e Speach To Text \u306e Whisper \u3092 C/C++ \u306b\u79fb\u690d\u3057\u305f\u3082\u306e\u304c\u3042\u308a\u307e\u3059\uff0e
\n\u65b0\u3057\u3044\u74b0\u5883\u3067\u4e45\u3057\u3076\u308a\u306b\u30bb\u30c3\u30c8\u30a2\u30c3\u30d7\u3057\u305f\u306e\u3067\u3059\u304c\uff0c\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u306b\u5931\u6557\u3057\u305f\u306e\u3067\u30e1\u30e2\u3057\u3066\u304a\u304d\u307e\u3059\uff0e

\n
\n

\n
\n\n
\n
\n

whisper.cpp \u3067\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u306b\u3092\u884c\u3046\u306e\u306b make \u3084 ./models/download-ggml-model.sh \u304c\u4f7f\u3048\u307e\u3059\uff0e

\n
\n
\n
base \u30e2\u30c7\u30eb\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u4f8b
\n
\n
$ make base\n$ bash ./models/download-ggml-model.sh base
\n
\n
\n
\n

\u30e2\u30c7\u30eb\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u4e2d\u306b\u5931\u6557\u3057\u3066\u3057\u307e\u3046\u3068\u518d\u5ea6\u5b9f\u884c\u3057\u3066\u3082\u3082\u3046\u30d5\u30a1\u30a4\u30eb\u3042\u308b\u3088\u3068\u8a00\u308f\u308c\u3059\u3050\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u7d42\u4e86\u3057\u3066\u3057\u307e\u3044\u307e\u3059\uff0e

\n
\n
\n
\n
Model base already exists. Skipping download.
\n
\n
\n
\n

\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u306f\u3069\u3046\u306a\u3063\u3066\u3044\u308b\u304b\u306a\u3068\u63a2\u3059\u3068\u6700\u8fd1\u306f models \u306e\u4e0b\u306b\u7f6e\u304f\u3088\u3046\u306a\u306e\u3067\u3053\u308c\u3092\u6d88\u3057\u3066\u518d\u5ea6\u5b9f\u884c\u3067\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u76f4\u305b\u308b\u3088\u3046\u3067\u3059\uff0c

\n
\n
\n
\n
$ rm models/ggml-base.bin
\n
\n
\n
\n

\u3053\u3061\u3089\u306b sha1(!) hash \u304c\u3042\u308b\u306e\u3067\u3053\u308c\u3068\u5408\u81f4\u3057\u306a\u3044\u5834\u5408\u6d88\u3059\u3088\u3046\u306b\u3059\u308b\u3068\u826f\u3055\u305d\u3046\u3067\u3059\uff0e

\n
\n
\n\n
\n
\n
\n
$ sha1sum models/ggml-base.bin\n465707469ff3a37a2b9b8d8f89f2f99de7299dac  models/ggml-base.bin
\n
\n
\n
\n

\u9014\u4e2d\u307e\u3067\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u304c\u9032\u3093\u3067\u3044\u308b\u5834\u5408\uff0cwget \u30b3\u30de\u30f3\u30c9\u306a\u3069\u3067\u7d9a\u304d\u304b\u3089\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3059\u308b\u3068\u8ee2\u9001\u6642\u9593\u304c\u5c11\u306a\u304f\u3066\u6e08\u307f\u307e\u3059\uff0e
\n\u30e2\u30c7\u30eb\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9URL \u306f\u4ee5\u4e0b\u306a\u3069\u304b\u3089\u78ba\u8a8d\u3067\u304d\u307e\u3059\uff0e

\n
\n
\n\n
\n
\n\n\n\n\n\n
\n
Note
\n		\nlarge \u304c\u6b32\u3057\u3044\u5834\u5408\u3053\u306e\u4e2d\u306b\u898b\u5f53\u305f\u308a\u307e\u305b\u3093\u304c\uff0c large-v3 \u304c\u305d\u308c\u306e\u3088\u3046\u3067\u3059\uff0e\n
\n
\n
\n
\u30d7\u30ed\u30b0\u30ec\u30b9\u306e + \u90e8\u5206\u306f\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u6e08\u3067\u30ec\u30b8\u30e5\u30fc\u30e0\u3057\u305f\u90e8\u5206\uff0c = \u304c\u4eca\u306e\u30bb\u30c3\u30b7\u30e7\u30f3\u3067\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u305f\u90e8\u5206
\n
\n
$ wget -c \"https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin?download=true\" \\\n-O models/ggml-medium.bin\n\n--snip--\n\nmoodels/ggml-medium.bin                      23%[+++++++++++++======>                                ] 341.88M   632KB/s
\n
\n
\n
\n

\u5b89\u5b9a\u3057\u305f\u74b0\u5883\u3060\u3068\u554f\u984c\u306a\u3044\u306e\u3067\u3057\u3087\u3046\u304c\u56de\u7dda\u3084\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u304c\u4e0d\u5b89\u5b9a\u3060\u3063\u305f\u308a(\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u3066\u3044\u308b\u306e\u3092\u5fd8\u308c\u3066\u30b5\u30b9\u30da\u30f3\u30c9\u3057\u3066\u3057\u307e\u3063\u305f\u308a)\u3059\u308b\u3068\u5d4c\u308b\u304b\u3082\u3057\u308c\u307e\u305b\u3093\uff0e

\n
\n
\n
\u74b0\u5883
\n
\n
$ git log -q -1\ncommit 6739eb83c3ca5cf40d24c6fe8442a761a1eb6248 (HEAD -> master, origin/master, origin/HEAD)\nAuthor: Georgi Gerganov <ggerganov@gmail.com>\nDate:   Sat Jul 27 20:35:04 2024 +0300\n\n    whisper : handle empty mel (#2324)\n$ lsb_release -dr\nDescription:    Debian GNU/Linux trixie/sid\nRelease:        n/a\n$ arch\nx86_64
\n
\n
\n", "content_text": "OpenAI \u306e Speach To Text \u306e Whisper \u3092 C/C++ \u306b\u79fb\u690d\u3057\u305f\u3082\u306e\u304c\u3042\u308a\u307e\u3059\uff0e\n\u65b0\u3057\u3044\u74b0\u5883\u3067\u4e45\u3057\u3076\u308a\u306b\u30bb\u30c3\u30c8\u30a2\u30c3\u30d7\u3057\u305f\u306e\u3067\u3059\u304c\uff0c\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u306b\u5931\u6557\u3057\u305f\u306e\u3067\u30e1\u30e2\u3057\u3066\u304a\u304d\u307e\u3059\uff0e\n\n\n\n\n\nopenai/whisper: Robust Speech Recognition via Large-Scale Weak Supervision\n\n\nggerganov/whisper.cpp: Port of OpenAI’s Whisper model in C/C++\n\n\n\n\nwhisper.cpp \u3067\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u306b\u3092\u884c\u3046\u306e\u306b make \u3084 ./models/download-ggml-model.sh \u304c\u4f7f\u3048\u307e\u3059\uff0e\n\n\nbase \u30e2\u30c7\u30eb\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u4f8b\n\n$ make base\n$ bash ./models/download-ggml-model.sh base\n\n\n\n\u30e2\u30c7\u30eb\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u4e2d\u306b\u5931\u6557\u3057\u3066\u3057\u307e\u3046\u3068\u518d\u5ea6\u5b9f\u884c\u3057\u3066\u3082\u3082\u3046\u30d5\u30a1\u30a4\u30eb\u3042\u308b\u3088\u3068\u8a00\u308f\u308c\u3059\u3050\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u7d42\u4e86\u3057\u3066\u3057\u307e\u3044\u307e\u3059\uff0e\n\n\n\nModel base already exists. Skipping download.\n\n\n\n\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u306f\u3069\u3046\u306a\u3063\u3066\u3044\u308b\u304b\u306a\u3068\u63a2\u3059\u3068\u6700\u8fd1\u306f models \u306e\u4e0b\u306b\u7f6e\u304f\u3088\u3046\u306a\u306e\u3067\u3053\u308c\u3092\u6d88\u3057\u3066\u518d\u5ea6\u5b9f\u884c\u3067\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u76f4\u305b\u308b\u3088\u3046\u3067\u3059\uff0c\n\n\n\n$ rm models/ggml-base.bin\n\n\n\n\u3053\u3061\u3089\u306b sha1(!) hash \u304c\u3042\u308b\u306e\u3067\u3053\u308c\u3068\u5408\u81f4\u3057\u306a\u3044\u5834\u5408\u6d88\u3059\u3088\u3046\u306b\u3059\u308b\u3068\u826f\u3055\u305d\u3046\u3067\u3059\uff0e\n\n\n\n\nggerganov/whisper.cpp \u00b7 Hugging Face\n\n\n\n\n\n$ sha1sum models/ggml-base.bin\n465707469ff3a37a2b9b8d8f89f2f99de7299dac models/ggml-base.bin\n\n\n\n\u9014\u4e2d\u307e\u3067\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u304c\u9032\u3093\u3067\u3044\u308b\u5834\u5408\uff0cwget \u30b3\u30de\u30f3\u30c9\u306a\u3069\u3067\u7d9a\u304d\u304b\u3089\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3059\u308b\u3068\u8ee2\u9001\u6642\u9593\u304c\u5c11\u306a\u304f\u3066\u6e08\u307f\u307e\u3059\uff0e\n\u30e2\u30c7\u30eb\u306e\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9URL \u306f\u4ee5\u4e0b\u306a\u3069\u304b\u3089\u78ba\u8a8d\u3067\u304d\u307e\u3059\uff0e\n\n\n\n\nggerganov/whisper.cpp at main\n\n\n\n\n\n\n\nNote\n\n\nlarge \u304c\u6b32\u3057\u3044\u5834\u5408\u3053\u306e\u4e2d\u306b\u898b\u5f53\u305f\u308a\u307e\u305b\u3093\u304c\uff0c large-v3 \u304c\u305d\u308c\u306e\u3088\u3046\u3067\u3059\uff0e\n\n\n\n\n\n\u30d7\u30ed\u30b0\u30ec\u30b9\u306e + \u90e8\u5206\u306f\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u6e08\u3067\u30ec\u30b8\u30e5\u30fc\u30e0\u3057\u305f\u90e8\u5206\uff0c = \u304c\u4eca\u306e\u30bb\u30c3\u30b7\u30e7\u30f3\u3067\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u305f\u90e8\u5206\n\n$ wget -c \"https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin?download=true\" \\\n-O models/ggml-medium.bin\n\n--snip--\n\nmoodels/ggml-medium.bin 23%[+++++++++++++======> ] 341.88M 632KB/s\n\n\n\n\u5b89\u5b9a\u3057\u305f\u74b0\u5883\u3060\u3068\u554f\u984c\u306a\u3044\u306e\u3067\u3057\u3087\u3046\u304c\u56de\u7dda\u3084\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u304c\u4e0d\u5b89\u5b9a\u3060\u3063\u305f\u308a(\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u3066\u3044\u308b\u306e\u3092\u5fd8\u308c\u3066\u30b5\u30b9\u30da\u30f3\u30c9\u3057\u3066\u3057\u307e\u3063\u305f\u308a)\u3059\u308b\u3068\u5d4c\u308b\u304b\u3082\u3057\u308c\u307e\u305b\u3093\uff0e\n\n\n\u74b0\u5883\n\n$ git log -q -1\ncommit 6739eb83c3ca5cf40d24c6fe8442a761a1eb6248 (HEAD -> master, origin/master, origin/HEAD)\nAuthor: Georgi Gerganov <ggerganov@gmail.com>\nDate: Sat Jul 27 20:35:04 2024 +0300\n\n whisper : handle empty mel (#2324)\n$ lsb_release -dr\nDescription: Debian GNU/Linux trixie/sid\nRelease: n/a\n$ arch\nx86_64", "date_published": "2024-08-02T06:35:38+09:00", "date_modified": "2024-08-02T06:38:05+09:00", "authors": [ { "name": "matoken", "url": "https://matoken.org/blog/author/matoken/", "avatar": "https://secure.gravatar.com/avatar/38f5f3b575c5eb45cda6aa659bca119ac7a5e16b46565e869d0030e3bd66981d?s=512&d=mm&r=g" } ], "author": { "name": "matoken", "url": "https://matoken.org/blog/author/matoken/", "avatar": "https://secure.gravatar.com/avatar/38f5f3b575c5eb45cda6aa659bca119ac7a5e16b46565e869d0030e3bd66981d?s=512&d=mm&r=g" }, "tags": [ "wget", "whisper", "whisper.cpp", "Debian", "Linux", "sid" ] } ] }