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

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

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

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