¿ªÆô×ó²à

Whisper ÓïÒôת»»ÎÄ×Ö¹¤¾ß£¬µÍÄڴ棬֧³Ö GPU

×ÊѶ
Whisper  ÓïÒôת»»ÎÄ×Ö¹¤¾ß£¬µÍÄڴ棬֧³Ö GPU


OpenAI µÄ Whisper ×Ô¶¯ÓïÒôʶ±ð (ASR) Ä£Ð͵ĸßÐÔÄÜ GPGPU ÍÆÀí
ʹÓà Whisper
ÔÚ Model Path ÀïÃæÑ¡ÔñÄãÏÂÔØµÄÄ£ÐÍ£¬È»ºóÑ¡Ôñ GPU ¾ÍÄܽøÈëÈí¼þÁË¡£
gui-load-model.png
Whisper - ±¾µØÓïÒôתÎÄ×Ö¹¤¾ß£¬Ö§³Ö GPU¡¢Ö§³ÖʵʱÓïÒôת»»[Windows] 1
È»ºóͨ¹ýÆÁÄ»ÔÊÐíת¼ÒôƵÎļþ¡£
gui-transcribe.png

»¹ÓÐÁíÒ»¸öÆÁÄ»ÔÊÐí´ÓÂó¿Ë·ç²¶»ñºÍת¼£¬»ò·­ÒëʵʱÒôƵÎļþ£¬Èç MP3 À´×ª»»ÎªÎÄ×ÖÁË¡£

gui-capture.png


»ùÓÚ DirectCompute µÄ¹©Ó¦É̲»¿ÉÖªµÄ GPGPU£»¸Ã¼¼ÊõµÄÁíÒ»¸öÃû³ÆÊÇ¡°Direct3D 11 ÖеļÆËã×ÅÉ«Æ÷¡±

´¿ C++ ʵÏÖ£¬³ýÁË»ù±¾µÄ OS ×é¼þÍâûÓÐÔËÐÐʱÒÀÀµÐÔ

±È OpenAI µÄʵÏÖ¿ìµÃ¶à¡£

ÔÚÎÒÅ䱸 GeForce 1080Ti GPU µÄÖÐÐĮ́ʽ¼ÆËã»úÉÏ£¬ ʹÓà PyTorch ºÍ CUDA ת¼3 ·Ö 24 ÃëµÄÑݽ²ÐèÒª 45 Ã룬µ«Ê¹ÓÃÎÒµÄʵÏÖºÍ DirectCompute Ö»ÐèÒª 19 Ãë¡£
ȤÎÅ£ºÕâÊÇ 9.63 GB µÄÔËÐÐʱÒÀÀµÏ¶ø 431 KBWhisper.dll

»ìºÏ F16 / F32 ¾«¶È£º×Ô D3D °æ±¾ 10.0 Æð£¬Windows ÐèÒª»º³åÇø µÄÖ§³ÖR16_FLOAT

ÄÚÖÃÐÔÄÜ·ÖÎöÆ÷£¬¿É²âÁ¿µ¥¸ö¼ÆËã×ÅÉ«Æ÷µÄÖ´ÐÐʱ¼ä

ÄÚ´æÊ¹ÓÃÂʵÍ

ÓÃÓÚÒôƵ´¦ÀíµÄýÌå»ù´¡£¬Ö§³Ö´ó¶àÊýÒôƵºÍÊÓÆµ¸ñʽ£¨Ogg Vorbis ³ýÍ⣩£¬ÒÔ¼°´ó¶àÊýÔÚ Windows ÉÏÔËÐеÄÒôƵ²¶»ñÉ豸£¨³ýÁËһЩרҵµÄ£¬ËüÃÇֻʵÏÖ ASIO API £©¡£

ÓÃÓÚÒôƵ²¶»ñµÄÓïÒô»î¶¯¼ì²â¡£

¸ÃʵÏÖ»ùÓÚMohammad Moattar ºÍ Mahdi Homayoonpoor ÓÚ2009 Äê·¢±íµÄÎÄÕ ¡°Ò»ÖÖ¼òµ¥µ«¸ßЧµÄʵʱÓïÒô»î¶¯¼ì²âËã·¨¡±¡£

https://www.researchgate.net/pub ... detection_algorithm


Ò×ÓÚʹÓÃµÄ COM ·ç¸ñ API£¬nuget ÉÏ¿ÉÓõĹßÓà C# °ü×°Æ÷

ÌṩԤ¹¹½¨µÄ¶þ½øÖÆÎļþ

ΨһÊÜÖ§³ÖµÄƽ̨ÊÇ 64 λ Windows¡£
Ó¦¸Ã¿ÉÒÔÔÚ Windows 8.1 »ò¸ü¸ß°æ±¾ÉÏÔËÐУ¬µ«ÎÒÖ»ÔÚ Windows 10 ÉϲâÊÔ¹ý¡£
¸Ã¿âÐèÒªÖ§³Ö Direct3D 11.0 µÄ GPU£¬ÕâÔÚ 2023 Äê¾ÍÒâζ×Å¡°ÈκÎÓ²¼þ GPU¡±¡£×îеIJ»Ö§³Ö D3D 11.0 µÄ GPU ÊÇ 2011 ÄêµÄ Intel Sandy Bridge¡£

Èí¼þÏÂÔØ£ºhttps://github.com/Const-me/Whisper
Ä£ÐÍ£ºhttps://huggingface.co/datasets/ggerganov/whisper.cpp/tree/main
ÊÕ²Ø
ËÍÔÞ
·ÖÏí

·¢±í»Ø¸´

µÇ¼ºó²ÅÄÜÆÀÂÛ

ÆÀÂÛÁÐ±í£¨1£©

ûÓÐÕûºÏ°æµÄÂð
2023-3-22 13:45:41 »Ø¸´

ÒôƵ·ÖÀà´óÈ«

+ ²é¿´È«²¿·ÖÀà