Prism

Personally
Responding,
Intelligent
Search
Matrix

Lua/Love2d based voice assistant.

Notes

The deepspeech module is complicated. My current setup has a lot of this stuff pre built, but it expects certain things: It expects you to have CUDA 10.1 and cuDNN v7.6.4.32 for CUDA 10.1 installed on your computer. It expects the system path variables for those to be on your computer (<install_location>/bin for both of them, and <install_location>/extras/CUPTI/lib64;<install_location>/include for CUDA). Obviously, since we're using CUDA, you have to have an Nvidia GPU. I have a GTX 1080. I don't know if it'll work OOB with other GPUs. You need to have the libdeepspeech.so file located next to either main.lua or the .exe if fused. (This is subject to change, I might try to setup some checks so this is done automatically.) A microphone. Deepspeech works with just raw audio files, but the version I distribute assumes you're doing real time speech to text, and so takes in audio from an input device.

You will almost defintely have to do all of this yourself, until I can figure out a nice way to distribute it: cmake from https://cmake.org/ native_client.amd64.gpu.win.tar.xz from https://github.com/Mozilla/DeepSpeech/releases/tag/v0.9.3 deepspeech-0.9.3-models.pbmm from https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.pbmm deepspeech-0.9.3-models.scorer from https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.scorer luajit from https://luajit.org/download/LuaJIT-2.1.0-beta3.zip CUDA 10.1 from https://developer.nvidia.com/cuda-10.1-download-archive-update2?target_os=Windows&target_arch=x86_64 (Note that you have to have an Nvidia Dev login, free to make) cuDNN v7.6.4.32 for CUDA 10.1 from https://developer.nvidia.com/compute/machine-learning/cudnn/secure/7.6.5.32/Production/10.1_20191031/cudnn-10.1-windows10-x64-v7.6.5.32.zip (Also need that same Nvidia Dev login) 7zip from https://www.7-zip.org/ lua-deepspeech from https://github.com/bjornbytes/lua-deepspeech (just download the whole project) Visual Studio Build Tools from https://visualstudio.microsoft.com/downloads/#build-tools-for-visual-studio-2019 (You're gunna need the C/C++ stuff, just download the whole thing)

1. Download each thing listed above. Install the things that can be installed. Extract the things that can be extracted using 7zip.
2. Set up sys PATH. You need: a. Path to cmake.exe (This you might need to do in user PATH. idk, that's how it's setup in mine.) b. <install_location>/bin for cuDNN c. <install_location>/bin for CUDA d. <install_location>/extras/CUPTI/lib64 for CUDA e. <install_location>/include for CUDA
3. RESTART YOUR COMPUTER. The PATH vars aren't set until a restart.
4. Open up the x64 build tools, and cd your way into the folder where you extracted the luajit stuff.
5. Drag drop the msvcbuild.bat onto the command prompt. It "should" run without any errors. If it has errors, fuck. Otherwise, you should get some files generated.
6. In the extracted folder for lua-deepspeech, open CMakeLists.text
7. Remove: a. include(FindLua) b. find_package(Lua REQUIRED)
8. Change: a. target_include_directories(lua-deepspeech PRIVATE "<location_of_src_folder>") b. target_link_libraries(lua-deepspeech PRIVATE "<location_of_src_folder>/lua51.lib")
9. Follow the build instruction on the lua-deepspeech page.
10. Point the -deepspeech_path to the extracted folder that came from the native_client tar
11. Go into the build folder, then Debug, and find lua-deepspeech.dll
12. Place the libdeepspeech.so file from the native_client tar into your project. If unfused, place it next to main.lua, otherwise place it next to the fused .exe.
13. Place the deepspeech-0.9.3-models.pbmm into a the local github project (mine, this one. Should have a folder)
14. Place the deepspeech-0.9.3-models.scorer next to it.
15. Put the lua-deepspeech.dll in the same folder.
16. If you're using my project, everything should now be hooked up.