WhisperX on Windows: Python, PyTorch and venv setup

This tutorial uses WhisperX to explain a Windows installation for local AI transcription. The walkthrough uses Python 3.10 and a separate venv so the program's dependencies do not conflict with other Python applications. The presenter recommends a folder near the drive root, with no spaces or Cyrillic characters in its path.

After activating the environment and updating pip, the presenter selects a PyTorch installation command for Windows and the intended compute platform. A CUDA availability check tests whether PyTorch detects the GPU. GPU compatibility remains a limitation: the tutorial does not provide a universal fix, and the presenter reports satisfactory CPU performance for his own short recordings.

The first WhisperX run downloads models; the presenter says subsequent runs use them locally. The example produces a transcript and an SRT subtitle file with timecodes. To keep model downloads off the system drive, the walkthrough sets HF_HOME and TORCH_HOME to folders on drive D and moves existing caches while preserving their folder hierarchy.

A BAT file activates the environment and accepts an audio or video file by drag and drop, placing results beside the input. Optional speaker diarization requires a Hugging Face account, acceptance of the model's agreement and a read-only access token. The presenter also supplies a batch-file variant without diarization for recordings that do not need speaker labels.