Windows installation instructions - update

[homelab-00]

Hi, even though it gave me a bit of trouble I installed it and it runs perfectly fine.
I've graciously decided to share my knowledge to the rest of the Windows script kiddies like me.
So:

1. 'uroman' was missing and needed to be installed separately via pip (not 100% certain about this one)
2. 'perl' needs to be installed on the system (installing it via strawberry perl worked fine on windows 10)
   I managed to trace down the 'perl' requirement and it's at L153 in text_utils.py located at "path\to\your\env\Lib\site-packages\ctc_forced_aligner"

Maybe consider updating the README as well.

[AIEnthusiastNN]

Hello,

Please guide me how to run it for windows. I am facing same bugs while installing dependencies and running the code locally.

[homelab-00]

I saw in another post that you have an RTX 3060 (same as me) so I'm using the CUDA version of the instructions. All right, from the top:

1. Install ffmpeg from gyan.dev. Then add the folder where ffmpeg.exe is located (e.g. C:\ffmpeg\bin) to the system variables PATH.
2. Install perl. It's a programming language like python. Strawberry perl worked fine for me.
3. Install Python 3.10.11. You can use later Python versions (e.g. 3.11, 3.12, etc), but 3.10.11 has given me no issues. There are later versions of 3.10 (e.g. 3.10.12) but there aren't binaries for windows and you have to build them yourself.
4. Create a Python virtual environment for your project. It's a good way to keep Python versions and dependencies separate for each project. To do that go to the root folder where you want to create the virtual environment, e.g. C:\Users\Bill\PurePython_envs. Open cmd at that location (or cd there) and run this command to create the Python virtual environment with venv: "C:\Program Files\Python310\python.exe" -m venv <your_venv_name>. The first part of this command specifies which Python version we want to use for this venv. Then cd into the venv folder by running cd "C:\Users\Bill\PurePython_envs\<your_venv_name>" and activate the venv by running scripts\activate. This command needs to be run from the environment folder (C:\Users\Bill\PurePython_envs\<your_venv_name> in this case). You're already in that folder at this point in the instructions but I'm including it so you know how to activate it in the future. If the venv has been activated correctly you should see the name of the venv before the location in cmd, e.g. (<your_venv_name>) C:\Users\Bill\PurePython_envs\<your_venv_name>.
5. Upgrade pip, setuptools and wheel by running: python -m pip install --upgrade pip setuptools wheel. I'm adding python -m before this command to make sure I'm using the correct Python environment.
6. Install CUDA Toolkit 12.4 and cuDNN (latest version is fine). There are newer versions of CUDA Toolkit but the latest version of pytorch (which is what we're using to GPU accelerate transcription) works with CUDA up to 12.4.
7. Install pytorch with CUDA 12.4 support. You can see the instructions for yourself here, but this is the command: python -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124
8. Install cython by running: python -m pip install cython
9. Download the zip of the entire whisper-diarization project and extract it somewhere, e.g. C:\Users\Bill\VSC_WorkingDirectory\whisper-diarization. cd into that folder from the same cmd terminal that has the venv active and we have been using for all the previous steps. Run this command to install whisper-diarization: C:\Users\Bill\PurePython_envs\<your_venv_name>\Scripts\python.exe -m pip install -c constraints.txt -r requirements.txt
10. Open the project folder (C:\Users\Bill\VSC_WorkingDirectory\whisper-diarization in this case) in VS Code and select the Python interpreter (by typing >Python: Select Interpreter in the command palette) of the venv we've created (in this case C:\Users\Bill\PurePython_envs\<your_venv_name>\scripts\python.exe).
11. Prepare your audio files by making sure that they're in .wav format, 16 KHz, mono (that's the format that whisper prefers (whisper is doing the transcription)). Audacity is a great free tool for that. Place the audio file in the project folder.
12. Finally to run the script open a terminal in the VS Code window we opened earlier and run: python diarize.py -a "your_audio_file.wav" --no-stem --whisper-model large-v3 --language en --device cuda. The resulting diarized transcriptions will appear as .txt and .srt files in the same location.

[AIEnthusiastNN]

Thank you so much. Let me try it. Will let you know. Stay blessed 🙏

[AIEnthusiastNN]

I saw in another post that you have an RTX 3060 (same as me) so I'm using the CUDA version of the instructions. All right, from the top:

1. Install ffmpeg from gyan.dev. Then add the folder where ffmpeg.exe is located (e.g. C:\ffmpeg\bin) to the system variables PATH.
2. Install perl. It's a programming language like python. Strawberry perl worked fine for me.
3. Install Python 3.10.11. You can use later Python versions (e.g. 3.11, 3.12, etc), but 3.10.11 has given me no issues. There are later versions of 3.10 (e.g. 3.10.12) but there aren't binaries for windows and you have to build them yourself.
4. Create a Python virtual environment for your project. It's a good way to keep Python versions and dependencies separate for each project. To do that go to the root folder where you want to create the virtual environment, e.g. C:\Users\Bill\PurePython_envs. Open cmd at that location (or cd there) and run this command to create the Python virtual environment with venv: "C:\Program Files\Python310\python.exe" -m venv <your_venv_name>. The first part of this command specifies which Python version we want to use for this venv. Then cd into the venv folder by running cd "C:\Users\Bill\PurePython_envs\<your_venv_name>" and activate the venv by running scripts\activate. This command needs to be run from the environment folder (C:\Users\Bill\PurePython_envs\<your_venv_name> in this case). You're already in that folder at this point in the instructions but I'm including it so you know how to activate it in the future. If the venv has been activated correctly you should see the name of the venv before the location in cmd, e.g. (<your_venv_name>) C:\Users\Bill\PurePython_envs\<your_venv_name>.
5. Upgrade pip, setuptools and wheel by running: python -m pip install --upgrade pip setuptools wheel. I'm adding python -m before this command to make sure I'm using the correct Python environment.
6. Install CUDA Toolkit 12.4 and cuDNN (latest version is fine). There are newer versions of CUDA Toolkit but the latest version of pytorch (which is what we're using to GPU accelerate transcription) works with CUDA up to 12.4.
7. Install pytorch with CUDA 12.4 support. You can see the instructions for yourself here, but this is the command: python -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124
8. Install uroman and cython by running: python -m pip install uroman cython
9. Download the zip of the entire whisper-diarization project and extract it somewhere, e.g. C:\Users\Bill\VSC_WorkingDirectory\whisper-diarization. cd into that folder from the same cmd terminal that has the venv active and we have been using for all the previous steps. Run this command to install whisper-diarization: C:\Users\Bill\PurePython_envs\<your_venv_name>\Scripts\python.exe -m pip install -c constraints.txt -r requirements.txt
10. Open the project folder (C:\Users\Bill\VSC_WorkingDirectory\whisper-diarization in this case) in VS Code and select the Python interpreter (by typing >Python: Select Interpreter in the command palette) of the venv we've created (in this case C:\Users\Bill\PurePython_envs\<your_venv_name>\scripts\python.exe).
11. Prepare your audio files by making sure that they're in .wav format, 16 KHz, mono (that's the format that whisper prefers (whisper is doing the transcription)). Audacity is a great free tool for that. Place the audio file in the project folder.
12. Finally to run the script open a terminal in the VS Code window we opened earlier and run: python diarize.py -a "your_audio_file.wav" --no-stem --whisper-model large-v3 --language en --device cuda. The resulting diarized transcriptions will appear as .txt and .srt files in the same location.

Thank you.... working now after installation of perl

[GuyRus]

Brilliant guide, thanks so much!

For anyone coming into this with a completely fresh windows install, you'll also need the Microsoft C++ build tools before install of whisper-diarization: https://visualstudio.microsoft.com/visual-cpp-build-tools/

[genewitch]

one note about your instructions regarding the venv

so what i do in projects like this:

#go into the folder whisper-diarisation will be in:
# like ~/apps/ 
git clone https://github.com/MahmoudAshraf97/whisper-diarization.git
cd whisper-diarization
python -m venv .
scripts/activate 

then go through the rest of your steps. When you want to run it again

cd ~/whisper-diarization
scripts/activate
python diarize.py <...>

now for windows, mostly the same but you need to be in powershell. if you have windows 11
you can open a cmd.exe terminal and press ctrl-shift+T to open a new tab, and that should be powershell.exe
if not you can Win+R "powershell.exe"
this is important because otherwise you can't tell if you're in the virtual environment or not;
also you can't do shell expansions with cmd.exe: $(dir *.mp3)

# root folder - in this example the "installed path" will be
# c:\whisper-diarization\ 
# note i am unsure if this exact command will work, i run everything in like d:\opt\whisper\
cd c:\
git clone https://github.com/MahmoudAshraf97/whisper-diarization.git
cd whisper-diarization
python -m venv .
#you need to run the following code snippet to allow powershell to *run* script files:
Set-ExecutionPolicy -ExecutionPolicy Bypass -Scope process
.\Scripts\Activate.ps1 

and then the rest. I'll check this again with fresh eyes in a bit, but this is what i do.

to go back into the venv after a reboot or whatever:

cd c:\whisper-diarization\
Set-ExecutionPolicy -ExecutionPolicy Bypass -Scope process
.\Scripts\Activate.ps1

[genewitch]

also windows has ffmpeg.exe, and iirc it's static single file executable, so you just copy it into your whisper-diarization folder.
This will let whisper open .webm, .mp4, .avi, .mp3, .aac, etc

https://www.gyan.dev/ffmpeg/builds/

[ilkz]

Hi. Have trouble with it on Win11 on step 9. Im get error with sentencepiece-0.2.0.tar.gz.
Have VS Build tools installed before.
Have Ninja binary, added to PATH.
Cant move to the next steps. Pls help with it...

[ilkz]

Solved by using Python 3.10.11. No problems appear after installing it.