20240202在Ubuntu20.04.6下使用whisper.cpp的CPU模式

20240202在Ubuntu20.04.6下使用whisper.cpp的CPU模式
2024/2/2 14:15

rootroot@rootroot-X99-Turbo:~/whisper.cpp$ ./main -l zh -osrt -m models/ggml-medium.bin chs.wav 
在纯CPU模式下，使用medium中等模型，7分钟的中文视频需要851829.69 ms，也就是852s，大概14分钟+。

https://github.com/ggerganov/whisper.cpp/tree/master/models
https://github.com/ggerganov/whisper.cpp
ggerganov/whisper.cpp

https://blog.csdn.net/aiyolo/article/details/129674728?share_token=2c48b804-37f6-43a8-9159-08b28147ad67
Whisper.cpp 编译使用
whisper.cpp 是牛人 ggerganov 对 openai 的 whisper 语音识别模型用 C++ 重新实现的项目，开源在 github 上，具有轻量、性能高，实用性强等特点。这篇文章主要记录在 windows 平台，如何使用该模型在本地端进行语音识别。
whisper.cpp 的开源地址在 ggerganov/whisper.cpp: Port of OpenAI’s Whisper model in C/C++ (github.com)，首先将项目下载在本地。
git clone https://github.com/ggerganov/whisper.cpp
whisper.cpp 项目里提供了几个现成的模型。建议下载 small 以上的模型，不然识别效果完全无法使用。

https://huggingface.co/ggerganov/whisper.cpp
ggerganov/whisper.cpp 
OpenAI's Whisper models converted to ggml format
Available models

Model    Disk    Mem    SHA
tiny    75 MB    ~390 MB    bd577a113a864445d4c299885e0cb97d4ba92b5f
tiny.en    75 MB    ~390 MB    c78c86eb1a8faa21b369bcd33207cc90d64ae9df
base    142 MB    ~500 MB    465707469ff3a37a2b9b8d8f89f2f99de7299dac
base.en    142 MB    ~500 MB    137c40403d78fd54d454da0f9bd998f78703390c
small    466 MB    ~1.0 GB    55356645c2b361a969dfd0ef2c5a50d530afd8d5
small.en    466 MB    ~1.0 GB    db8a495a91d927739e50b3fc1cc4c6b8f6c2d022
medium    1.5 GB    ~2.6 GB    fd9727b6e1217c2f614f9b698455c4ffd82463b4
medium.en    1.5 GB    ~2.6 GB    8c30f0e44ce9560643ebd10bbe50cd20eafd3723
large-v1    2.9 GB    ~4.7 GB    b1caaf735c4cc1429223d5a74f0f4d0b9b59a299
large-v2    2.9 GB    ~4.7 GB    0f4c8e34f21cf1a914c59d8b3ce882345ad349d6
large    2.9 GB    ~4.7 GB    ad82bf6a9043ceed055076d0fd39f5f186ff8062
note: large corresponds to the latest Large v3 model

For more information, visit:

https://github.com/ggerganov/whisper.cpp/tree/master/models
https://huggingface.co/ggerganov/whisper.cpp/tree/main

【结论：在Ubuntu20.04.6下，确认large模式识别7分钟中文视频，需要356447.78 ms，也就是356.5秒，需要大概5分钟！效率太差！】
前提条件，可以通过技术手段上外网！^_
首先你要有一张NVIDIA的显卡，比如我用的PDD拼多多的二手GTX1080显卡。【并且极其可能是矿卡！】800￥
2、请正确安装好NVIDIA最新的545版本的驱动程序和CUDA、cuDNN。
2、安装Torch
3、配置whisper

https://github.com/ggerganov/whisper.cpp
https://www.toutiao.com/article/7276732434920653312/?app=news_article&timestamp=1706802934&use_new_style=1&req_id=2024020123553463D3509B1706BC79D479&group_id=7276732434920653312&tt_from=mobile_qq&utm_source=mobile_qq&utm_medium=toutiao_android&utm_campaign=client_share&share_token=7bcb7488-a03d-4291-96fb-d0835ac76cca&source=m_redirect
https://www.toutiao.com/article/7276732434920653312/
OpenAI的whisper的c/c++ 版本体验

首先下载代码,注：我的OS环境是Ubuntu20.04.6。
git clone https://github.com/ggerganov/whisper.cpp

下载成功后进入项目目录：
cd whisper.cpp

执行如下脚本命令下载模型，这里选择的base 版本，我们先来测试英语识别：
bash ./models/download-ggml-model.sh base.en
但是尝试了几次都无法下载成功，报错消息如下：

网上search 了一下，找到可提供下载的链接：
https://github.com/ggerganov/whisper.cpp/tree/master/models
https://huggingface.co/ggerganov/whisper.cpp/tree/main
我选择下载全部35个文件！

下载成功后将模型文件copy 到项目中的models目录：
cp ~/Downloads/ggml-base.en.gin /home/havelet/ai/whisper.cpp/models

接下来执行如下编译命令：
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ make
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 

执行结果如下：
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ make
I whisper.cpp build info: 
I UNAME_S:  Linux
I UNAME_P:  x86_64
I UNAME_M:  x86_64
I CFLAGS:   -I.              -O3 -DNDEBUG -std=c11   -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3
I CXXFLAGS: -I. -I./examples -O3 -DNDEBUG -std=c++11 -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3
I LDFLAGS:  
I CC:       cc (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0
I CXX:      g++ (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0

cc  -I.              -O3 -DNDEBUG -std=c11   -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3   -c ggml.c -o ggml.o
cc  -I.              -O3 -DNDEBUG -std=c11   -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3   -c ggml-alloc.c -o ggml-alloc.o
cc  -I.              -O3 -DNDEBUG -std=c11   -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3   -c ggml-backend.c -o ggml-backend.o
cc  -I.              -O3 -DNDEBUG -std=c11   -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3   -c ggml-quants.c -o ggml-quants.o
g++ -I. -I./examples -O3 -DNDEBUG -std=c++11 -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3 -c whisper.cpp -o whisper.o
g++ -I. -I./examples -O3 -DNDEBUG -std=c++11 -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3 examples/main/main.cpp examples/common.cpp examples/common-ggml.cpp ggml.o ggml-alloc.o ggml-backend.o ggml-quants.o whisper.o -o main 
./main -h

usage: ./main [options] file0.wav file1.wav ...

options:
  -h,        --help              [default] show this help message and exit
  -t N,      --threads N         [4      ] number of threads to use during computation
  -p N,      --processors N      [1      ] number of processors to use during computation
  -ot N,     --offset-t N        [0      ] time offset in milliseconds
  -on N,     --offset-n N        [0      ] segment index offset
  -d  N,     --duration N        [0      ] duration of audio to process in milliseconds
  -mc N,     --max-context N     [-1     ] maximum number of text context tokens to store
  -ml N,     --max-len N         [0      ] maximum segment length in characters
  -sow,      --split-on-word     [false  ] split on word rather than on token
  -bo N,     --best-of N         [5      ] number of best candidates to keep
  -bs N,     --beam-size N       [5      ] beam size for beam search
  -wt N,     --word-thold N      [0.01   ] word timestamp probability threshold
  -et N,     --entropy-thold N   [2.40   ] entropy threshold for decoder fail
  -lpt N,    --logprob-thold N   [-1.00  ] log probability threshold for decoder fail
  -debug,    --debug-mode        [false  ] enable debug mode (eg. dump log_mel)
  -tr,       --translate         [false  ] translate from source language to english
  -di,       --diarize           [false  ] stereo audio diarization
  -tdrz,     --tinydiarize       [false  ] enable tinydiarize (requires a tdrz model)
  -nf,       --no-fallback       [false  ] do not use temperature fallback while decoding
  -otxt,     --output-txt        [false  ] output result in a text file
  -ovtt,     --output-vtt        [false  ] output result in a vtt file
  -osrt,     --output-srt        [false  ] output result in a srt file
  -olrc,     --output-lrc        [false  ] output result in a lrc file
  -owts,     --output-words      [false  ] output script for generating karaoke video
  -fp,       --font-path         [/System/Library/Fonts/Supplemental/Courier New Bold.ttf] path to a monospace font for karaoke video
  -ocsv,     --output-csv        [false  ] output result in a CSV file
  -oj,       --output-json       [false  ] output result in a JSON file
  -ojf,      --output-json-full  [false  ] include more information in the JSON file
  -of FNAME, --output-file FNAME [       ] output file path (without file extension)
  -np,       --no-prints         [false  ] do not print anything other than the results
  -ps,       --print-special     [false  ] print special tokens
  -pc,       --print-colors      [false  ] print colors
  -pp,       --print-progress    [false  ] print progress
  -nt,       --no-timestamps     [false  ] do not print timestamps
  -l LANG,   --language LANG     [en     ] spoken language ('auto' for auto-detect)
  -dl,       --detect-language   [false  ] exit after automatically detecting language
             --prompt PROMPT     [       ] initial prompt
  -m FNAME,  --model FNAME       [models/ggml-base.en.bin] model path
  -f FNAME,  --file FNAME        [       ] input WAV file path
  -oved D,   --ov-e-device DNAME [CPU    ] the OpenVINO device used for encode inference
  -ls,       --log-score         [false  ] log best decoder scores of tokens
  -ng,       --no-gpu            [false  ] disable GPU

g++ -I. -I./examples -O3 -DNDEBUG -std=c++11 -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3 examples/bench/bench.cpp ggml.o ggml-alloc.o ggml-backend.o ggml-quants.o whisper.o -o bench 
g++ -I. -I./examples -O3 -DNDEBUG -std=c++11 -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3 examples/quantize/quantize.cpp examples/common.cpp examples/common-ggml.cpp ggml.o ggml-alloc.o ggml-backend.o ggml-quants.o whisper.o -o quantize 
g++ -I. -I./examples -O3 -DNDEBUG -std=c++11 -fPIC -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -pthread -mavx -mavx2 -mfma -mf16c -msse3 -mssse3 examples/server/server.cpp examples/common.cpp examples/common-ggml.cpp ggml.o ggml-alloc.o ggml-backend.o ggml-quants.o whisper.o -o server  
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 

编译成功后，则可以执行测试程序，首先执行自带测试音频：【英文】
./main -f samples/jfk.wav
执行结果如下，我们可看到识别结果正确：

rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ ./main -f samples/jfk.wav
whisper_init_from_file_with_params_no_state: loading model from 'models/ggml-base.en.bin'
whisper_model_load: loading model
whisper_model_load: n_vocab       = 51864
whisper_model_load: n_audio_ctx   = 1500
whisper_model_load: n_audio_state = 512
whisper_model_load: n_audio_head  = 8
whisper_model_load: n_audio_layer = 6
whisper_model_load: n_text_ctx    = 448
whisper_model_load: n_text_state  = 512
whisper_model_load: n_text_head   = 8
whisper_model_load: n_text_layer  = 6
whisper_model_load: n_mels        = 80
whisper_model_load: ftype         = 1
whisper_model_load: qntvr         = 0
whisper_model_load: type          = 2 (base)
whisper_model_load: adding 1607 extra tokens
whisper_model_load: n_langs       = 99
whisper_model_load:      CPU total size =   147.46 MB (1 buffers)
whisper_model_load: model size    =  147.37 MB
whisper_init_state: kv self size  =   16.52 MB
whisper_init_state: kv cross size =   18.43 MB
whisper_init_state: compute buffer (conv)   =   16.17 MB
whisper_init_state: compute buffer (encode) =   94.42 MB
whisper_init_state: compute buffer (cross)  =    5.08 MB
whisper_init_state: compute buffer (decode) =  105.96 MB

system_info: n_threads = 4 / 36 | AVX = 1 | AVX2 = 1 | AVX512 = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | METAL = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | CUDA = 0 | COREML = 0 | OPENVINO = 0 | 

main: processing 'samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, 5 beams + best of 5, lang = en, task = transcribe, timestamps = 1 ...

[00:00:00.000 --> 00:00:11.000]   And so my fellow Americans, ask not what your country can do for you, ask what you can do for your country.

whisper_print_timings:     load time =   148.05 ms
whisper_print_timings:     fallbacks =   0 p /   0 h
whisper_print_timings:      mel time =    26.18 ms
whisper_print_timings:   sample time =    85.77 ms /   131 runs (    0.65 ms per run)
whisper_print_timings:   encode time =  1716.83 ms /     1 runs ( 1716.83 ms per run)
whisper_print_timings:   decode time =    18.81 ms /     2 runs (    9.40 ms per run)
whisper_print_timings:   batchd time =   399.91 ms /   125 runs (    3.20 ms per run)
whisper_print_timings:   prompt time =     0.00 ms /     1 runs (    0.00 ms per run)
whisper_print_timings:    total time =  2404.17 ms
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ ./main -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 16
whisper_init_from_file_with_params_no_state: loading model from './models/ggml-base.en.bin'
whisper_model_load: loading model
whisper_model_load: n_vocab       = 51864
whisper_model_load: n_audio_ctx   = 1500
whisper_model_load: n_audio_state = 512
whisper_model_load: n_audio_head  = 8
whisper_model_load: n_audio_layer = 6
whisper_model_load: n_text_ctx    = 448
whisper_model_load: n_text_state  = 512
whisper_model_load: n_text_head   = 8
whisper_model_load: n_text_layer  = 6
whisper_model_load: n_mels        = 80
whisper_model_load: ftype         = 1
whisper_model_load: qntvr         = 0
whisper_model_load: type          = 2 (base)
whisper_model_load: adding 1607 extra tokens
whisper_model_load: n_langs       = 99
whisper_model_load:      CPU total size =   147.46 MB (1 buffers)
whisper_model_load: model size    =  147.37 MB
whisper_init_state: kv self size  =   16.52 MB
whisper_init_state: kv cross size =   18.43 MB
whisper_init_state: compute buffer (conv)   =   16.17 MB
whisper_init_state: compute buffer (encode) =   94.42 MB
whisper_init_state: compute buffer (cross)  =    5.08 MB
whisper_init_state: compute buffer (decode) =  105.96 MB

system_info: n_threads = 4 / 36 | AVX = 1 | AVX2 = 1 | AVX512 = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | METAL = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | CUDA = 0 | COREML = 0 | OPENVINO = 0 | 

main: processing './samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, 5 beams + best of 5, lang = en, task = transcribe, timestamps = 1 ...

[00:00:00.000 --> 00:00:00.850]   And so my
[00:00:00.850 --> 00:00:01.590]   fellow
[00:00:01.590 --> 00:00:04.140]   Americans, ask
[00:00:04.140 --> 00:00:05.740]   not what your
[00:00:05.740 --> 00:00:07.000]   country can do
[00:00:07.000 --> 00:00:08.370]   for you, ask
[00:00:08.370 --> 00:00:09.440]   what you can do
[00:00:09.440 --> 00:00:10.020]   for your
[00:00:10.020 --> 00:00:11.000]   country.

whisper_print_timings:     load time =   146.52 ms
whisper_print_timings:     fallbacks =   0 p /   0 h
whisper_print_timings:      mel time =    27.35 ms
whisper_print_timings:   sample time =    87.86 ms /   131 runs (    0.67 ms per run)
whisper_print_timings:   encode time =  1829.62 ms /     1 runs ( 1829.62 ms per run)
whisper_print_timings:   decode time =    20.76 ms /     2 runs (   10.38 ms per run)
whisper_print_timings:   batchd time =   401.02 ms /   125 runs (    3.21 ms per run)
whisper_print_timings:   prompt time =     0.00 ms /     1 runs (    0.00 ms per run)
whisper_print_timings:    total time =  2550.71 ms
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 

rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ ffmpeg -i chs.mp4 -ar 16000 -ac 1 -c:a pcm_s16le chs.wav
ffmpeg version 4.2.7-0ubuntu0.1 Copyright (c) 2000-2022 the FFmpeg developers
  built with gcc 9 (Ubuntu 9.4.0-1ubuntu1~20.04.1)
  configuration: --prefix=/usr --extra-version=0ubuntu0.1 --toolchain=hardened --libdir=/usr/lib/x86_64-linux-gnu --incdir=/usr/include/x86_64-linux-gnu --arch=amd64 --enable-gpl --disable-stripping --enable-avresample --disable-filter=resample --enable-avisynth --enable-gnutls --enable-ladspa --enable-libaom --enable-libass --enable-libbluray --enable-libbs2b --enable-libcaca --enable-libcdio --enable-libcodec2 --enable-libflite --enable-libfontconfig --enable-libfreetype --enable-libfribidi --enable-libgme --enable-libgsm --enable-libjack --enable-libmp3lame --enable-libmysofa --enable-libopenjpeg --enable-libopenmpt --enable-libopus --enable-libpulse --enable-librsvg --enable-librubberband --enable-libshine --enable-libsnappy --enable-libsoxr --enable-libspeex --enable-libssh --enable-libtheora --enable-libtwolame --enable-libvidstab --enable-libvorbis --enable-libvpx --enable-libwavpack --enable-libwebp --enable-libx265 --enable-libxml2 --enable-libxvid --enable-libzmq --enable-libzvbi --enable-lv2 --enable-omx --enable-openal --enable-opencl --enable-opengl --enable-sdl2 --enable-libdc1394 --enable-libdrm --enable-libiec61883 --enable-nvenc --enable-chromaprint --enable-frei0r --enable-libx264 --enable-shared
  libavutil      56. 31.100 / 56. 31.100
  libavcodec     58. 54.100 / 58. 54.100
  libavformat    58. 29.100 / 58. 29.100
  libavdevice    58.  8.100 / 58.  8.100
  libavfilter     7. 57.100 /  7. 57.100
  libavresample   4.  0.  0 /  4.  0.  0
  libswscale      5.  5.100 /  5.  5.100
  libswresample   3.  5.100 /  3.  5.100
  libpostproc    55.  5.100 / 55.  5.100
Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'chs.mp4':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2avc1mp41iso5
    comment         : vid:v0d004g10000cmbmsrjc77ubc8r79ssg
    encoder         : Lavf58.76.100
  Duration: 00:07:01.78, start: -0.042667, bitrate: 65 kb/s
    Stream #0:0(und): Audio: aac (LC) (mp4a / 0x6134706D), 48000 Hz, stereo, fltp, 64 kb/s (default)
    Metadata:
      handler_name    : Bento4 Sound Handler
Stream mapping:
  Stream #0:0 -> #0:0 (aac (native) -> pcm_s16le (native))
Press [q] to stop, [?] for help
Output #0, wav, to 'chs.wav':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2avc1mp41iso5
    ICMT            : vid:v0d004g10000cmbmsrjc77ubc8r79ssg
    ISFT            : Lavf58.29.100
    Stream #0:0(und): Audio: pcm_s16le ([1][0][0][0] / 0x0001), 16000 Hz, mono, s16, 256 kb/s (default)
    Metadata:
      handler_name    : Bento4 Sound Handler
      encoder         : Lavc58.54.100 pcm_s16le
size=   13181kB time=00:07:01.78 bitrate= 256.0kbits/s speed= 645x    
video:0kB audio:13181kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.000919%
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ ll *.wav
-rw-rw-r-- 1 rootroot rootroot 13497126 2月   2 17:26 chs.wav
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ ./main -l zh -osrt -m models/ggml-medium.bin chs.wav 
whisper_init_from_file_with_params_no_state: loading model from 'models/ggml-medium.bin'
whisper_model_load: loading model
whisper_model_load: n_vocab       = 51865
whisper_model_load: n_audio_ctx   = 1500
whisper_model_load: n_audio_state = 1024
whisper_model_load: n_audio_head  = 16
whisper_model_load: n_audio_layer = 24
whisper_model_load: n_text_ctx    = 448
whisper_model_load: n_text_state  = 1024
whisper_model_load: n_text_head   = 16
whisper_model_load: n_text_layer  = 24
whisper_model_load: n_mels        = 80
whisper_model_load: ftype         = 1
whisper_model_load: qntvr         = 0
whisper_model_load: type          = 4 (medium)
whisper_model_load: adding 1608 extra tokens
whisper_model_load: n_langs       = 99
whisper_model_load:      CPU total size =  1533.52 MB (2 buffers)
whisper_model_load: model size    = 1533.14 MB
whisper_init_state: kv self size  =  132.12 MB
whisper_init_state: kv cross size =  147.46 MB
whisper_init_state: compute buffer (conv)   =   28.00 MB
whisper_init_state: compute buffer (encode) =  187.14 MB
whisper_init_state: compute buffer (cross)  =    8.46 MB
whisper_init_state: compute buffer (decode) =  107.98 MB

system_info: n_threads = 4 / 36 | AVX = 1 | AVX2 = 1 | AVX512 = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | METAL = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | CUDA = 0 | COREML = 0 | OPENVINO = 0 | 

main: processing 'chs.wav' (6748501 samples, 421.8 sec), 4 threads, 1 processors, 5 beams + best of 5, lang = zh, task = transcribe, timestamps = 1 ...

[00:00:00.000 --> 00:00:01.520]  前段时间有个巨石横虎
[00:00:01.520 --> 00:00:03.040]  某某是男人最好的义妹
[00:00:03.040 --> 00:00:04.800]  这里的某某可以替换为减肥
[00:00:04.800 --> 00:00:07.760]  长发 西装 考研 书唱 永洁无间等等等等
[00:00:07.760 --> 00:00:09.280]  我听到最新的一个说法是
[00:00:09.280 --> 00:00:12.000]  微分碎盖加口罩加半框眼镜加春风衣
[00:00:12.000 --> 00:00:13.360]  等于男人最好的义妹

[00:06:28.920 --> 00:06:32.920]  何况骆驼也多多少少有一些功能性 毕竟它再怎么样还是个冲锋衣
[00:06:32.920 --> 00:06:35.920]  理解了这个事情就很容易分辨什么是智商税的
[00:06:35.920 --> 00:06:40.920]  那些向你灌输非某个品牌不用 告诉你某个需求只有某个产品才能满足
[00:06:40.920 --> 00:06:45.920]  某个品牌就是某个品类绝对的鄙视链顶端 这类营销的智商税含量必然是很高的
[00:06:46.920 --> 00:06:50.920]  它的目的是剥夺你选择的权利 让你主动放弃比价和寻找平梯的想法
[00:06:50.920 --> 00:06:55.920]  从而避免与其他品牌竞争 而没有竞争的市场才是智商税含量最高的市场
[00:06:55.920 --> 00:06:59.920]  消费 商业洞穴 尽在IC实验室 我是馆长 我们下期再见
[00:06:59.920 --> 00:07:01.920]  謝謝收看!

output_srt: saving output to 'chs.wav.srt'

whisper_print_timings:     load time =   699.72 ms
whisper_print_timings:     fallbacks =   1 p /   0 h
whisper_print_timings:      mel time =   435.93 ms
whisper_print_timings:   sample time = 13399.07 ms / 19082 runs (    0.70 ms per run)
whisper_print_timings:   encode time = 371019.09 ms /    19 runs (19527.32 ms per run)
whisper_print_timings:   decode time =   895.08 ms /    16 runs (   55.94 ms per run)
whisper_print_timings:   batchd time = 409084.94 ms / 19045 runs (   21.48 ms per run)
whisper_print_timings:   prompt time = 56250.60 ms /  3866 runs (   14.55 ms per run)
whisper_print_timings:    total time = 851829.69 ms
rootroot@rootroot-X99-Turbo:~/whisper.cpp$ 

中等模式下的显存占用：

参考资料：
https://www.toutiao.com/article/7225218604160418338/?app=news_article&timestamp=1706803458&use_new_style=1&req_id=2024020200041726E9258609E554857D25&group_id=7225218604160418338&tt_from=mobile_qq&utm_source=mobile_qq&utm_medium=toutiao_android&utm_campaign=client_share&share_token=37e094d5-29b8-4d14-87bb-241cdc28b0ea&source=m_redirect
AI浪潮下的12大开源神器介绍
原创2023-04-23 20:33·IT小熊实验室丶

https://blog.csdn.net/sinat_18131557/article/details/130950719?share_token=25ca6bb5-8450-472c-9228-abc8c6ce74d8
whisper.cpp在Windows VS的编译
sinat_18131557 于 2023-05-30 16:03:53 发布

https://www.toutiao.com/article/7283079784329052726/?app=news_article&timestamp=1706803297&use_new_style=1&req_id=20240202000137411974769524167990E0&group_id=7283079784329052726&tt_from=mobile_qq&utm_source=mobile_qq&utm_medium=toutiao_android&utm_campaign=client_share&share_token=b7961b29-d87a-4b6c-bb8e-c7c213388390&source=m_redirect
【往期回顾】Github开源项目月刊精选-2023年8月
原创2023-09-27 08:30·Github推荐官

https://blog.csdn.net/weixin_45533131/article/details/132817683?share_token=72d8a161-4d49-4795-ad21-2ce5e2e4b197
在Linux(Centos7)上编译whisper.cpp的详细教程

https://github.com/Const-me/Whisper/releases
https://www.cnblogs.com/jike9527/p/17545484.html?share_token=5af4092d-5b67-4e52-8231-0ae220fd2185
使用whisper批量生成字幕(whisper.cpp)

https://blog.csdn.net/u012234115/article/details/134668510?share_token=e3835a0d-ac3b-4c86-9e32-e79ec85cddbe
开源C++智能语音识别库whisper.cpp开发使用入门

https://www.toutiao.com/article/7276732434920653312/?app=news_article&timestamp=1706802934&use_new_style=1&req_id=2024020123553463D3509B1706BC79D479&group_id=7276732434920653312&tt_from=mobile_qq&utm_source=mobile_qq&utm_medium=toutiao_android&utm_campaign=client_share&share_token=7bcb7488-a03d-4291-96fb-d0835ac76cca&source=m_redirect
OpenAI的whisper的c/c++ 版本体验
首先下载代码,注：我的OS环境是ubuntu 18.04。

https://post.smzdm.com/p/a3052kz7/?share_token=d4057cba-adb0-4c91-8a8b-d8a7adcf4087
显卡怎么玩 篇三：音频转字幕神器whisper升级版，whisper-webui使用教程

https://www.toutiao.com/article/7311876528407921162/?app=news_article&timestamp=1706801102&use_new_style=1&req_id=20240201232501647517150775FC7AD89A&group_id=7311876528407921162&tt_from=mobile_qq&utm_source=mobile_qq&utm_medium=toutiao_android&utm_campaign=client_share&share_token=dfa1976e-9422-49d2-a73b-6453becea90c&source=m_redirect
2023 AI 界7个最火的 Text-to-Video 模型

动画
https://www.toutiao.com/article/7312473532829745700/?app=news_article&timestamp=1706801052&use_new_style=1&req_id=2024020123241265D9BE3F954EB979A010&group_id=7312473532829745700&tt_from=mobile_qq&utm_source=mobile_qq&utm_medium=toutiao_android&utm_campaign=client_share&share_token=ca5d0d2a-2d9b-4959-b5c0-3dd869555240&source=m_redirect
推荐5款本周 超火 的开源AI项目
原创2023-12-15 07:32·程序员梓羽同学

https://blog.csdn.net/chenlu5201314/article/details/131156770?share_token=b8796ff0-44f8-471a-af6d-c1bc7ca57002
【开源工具】使用Whisper提取视频、语音的字幕
1、下载安装包Assets\WhisperDesktop.zip

https://www.toutiao.com/article/7222852915286016544/?app=news_article&timestamp=1706460752&use_new_style=1&req_id=2024012900523164164830D4E1ECF3CCE2&group_id=7222852915286016544&tt_from=mobile_qq&utm_source=mobile_qq&utm_medium=toutiao_android&utm_campaign=client_share&share_token=9bc8621f-b3b1-4f49-ae20-5214c1254515&source=m_redirect
从零开始，手把手教本地部署Stable Diffusion AI绘画 V3版 (Win最新)
原创2023-04-17 11:23·觉悟之坡

https://blog.csdn.net/S_eashell/article/details/135258411?share_token=f998e896-6dff-4fd4-8df2-c6aae132e95c
98秒转录2.5小时音频，最强音频转文字软件insanely-fast-whisper下载部署
老艾的AI世界 已于 2024-01-05 20:20:51 修改