Has someone do inference (denoise an audio) successfully?

[betegon]

I have been digging into the code but I haven't been able to make it work, what I want to achieve is just to denoise files.

What flow have you follow to achieve it?

[his0car]

I'm not sure why you put the :/ emoticon, whether it is the question that is ":/" or you couldn't make it work either. Anyway, for us it did actually work (to run inference, we haven't tried training), so we are writing out some steps and configs to help others:

We use pycharm on Windows 10.

• Cuda version: 10.1 (Update 2!, 1 won't work)
• Microsoft Visual Studio 2019 Community Edition with the C++ package (can't remember which one)
• Put C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30037\bin\Hostx64\x64 and C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30037\bin\Hostx64\x64\cl.exe in the Environment Variables in PATH (System Variables)
• These are the currently installed packages in pycharm
• We used Conda as well as PIP to install these (and we are aware that mixing these isn't good practice) and most of them installed themselves as part of other packages

blas | 1.0 |   ca-certificates | 2021.7.5 |   certifi | 2020.6.20 |   cffi | 1.14.6 |   cython | 0.25.2 |   h5py | 2.7.0 |   icc_rt | 2019.0.0 |   intel-openmp | 2021.2.0 |   keras | 1.2.0 |   libgpuarray | 0.6.9 |   libpython | 2.1 |   m2w64-binutils | 2.25.1 |   m2w64-bzip2 | 1.0.6 |   m2w64-crt-git | 5.0.0.4636.2595836 |   m2w64-gcc | 5.3.0 |   m2w64-gcc-ada | 5.3.0 |   m2w64-gcc-fortran | 5.3.0 |   m2w64-gcc-libgfortran | 5.3.0 |   m2w64-gcc-libs | 5.3.0 |   m2w64-gcc-libs-core | 5.3.0 |   m2w64-gcc-objc | 5.3.0 |   m2w64-gmp | 6.1.0 |   m2w64-headers-git | 5.0.0.4636.c0ad18a |   m2w64-isl | 0.16.1 |   m2w64-libiconv | 1.14 |   m2w64-libmangle-git | 5.0.0.4509.2e5a9a2 |   m2w64-libwinpthread-git | 5.0.0.4634.697f757 |   m2w64-make | 4.1.2351.a80a8b8 |   m2w64-mpc | 1.0.3 |   m2w64-mpfr | 3.1.4 |   m2w64-pkg-config | 0.29.1 |   m2w64-toolchain | 5.3.0 |   m2w64-tools-git | 5.0.0.4592.90b8472 |   m2w64-windows-default-manifest | 6.4 |   m2w64-winpthreads-git | 5.0.0.4634.697f757 |   m2w64-zlib | 1.2.8 |   mako | 1.1.4 |   markupsafe | 1.1.1 |   mkl | 2020.2 |   mkl-service | 2.3.0 |   msys2-conda-epoch | 20160418 |   numpy | 1.16.6 |   pip | 19.3.1 |   pycparser | 2.20 |   pygpu | 0.6.9 |   python | 2.7.18 |   pyyaml | 5.4.1 |   scipy | 0.19.0 |   setuptools | 44.0.0 |   six | 1.16.0 |   soundfile | 0.9.0 |   sqlite | 3.30.1 |   theano | 1.0.5 |   tqdm | 4.11.0 |   vc | 9 |   vs2008_runtime | 9.00.30729.1 |   wheel | 0.36.2 |   wincertstore | 0.2 |  

• In the keras.json config file, which is located at "C:\Users\My_User_Name.keras" we changed the line from "backend": "tensorflow" to "backend": "theano"
• In the project itself we made sure that the loaded checkpoint pointed to the correct path
• To run on gpu etc. of course use the pycharm "Run..." -> "edit configuration"

[Thanoon12k]

not worked for me