hey! this is a static git repo viewer. commit diffs, file history, older commits and other revisions won't load without javascript.

you are still able to view source files on branch tips, but the experience is degraded.

yum-archive/TaSTT-Whisper

High-performance GPGPU inference of OpenAI's Whisper automatic speech recognition (ASR) model

git clone https://git.yummers.dev/yum-archive/TaSTT-Whisper

yumFix audio normalization59297502023-04-04

files

API/
CPU/
D3D/
Hybrid/
MF/
ML/
Utils/
Whisper/
WhisperCLI/
source/
source.compat/
DllMain.cpp	707 B
Readme.txt	525 B
Resource.rc	5.1 KiB
Whisper.vcxproj	17.8 KiB
Whisper.vcxproj.filters	9.9 KiB
misc.natvis	2.6 KiB
modelFactory.cpp	731 B
modelFactory.h	304 B
resource.h	388 B
stdafx.cpp	19 B
stdafx.h	1.6 KiB
whisper.def	153 B
whisperCom.cpp	37.4 KiB

This C++ project builds a DLL which actually does the heavy lifting of this project.

It implements the ML model, handles multimedia files with Media Foundation, captures audio (also with MF), does voice activity detection (custom code running on CPU), and a few smaller things.

The code requires C++/20, and only tested with Visual Studio 2022.

When running pure GPGPU model, the DLL requires SSE 4.1 instruction set.

When running a hybrid model, the DLL requires AVX1, FMA3, F16C, and BMI1 instruction set extensions.