WhisperX Transcription GUI

A GUI for transcribing audio and video files using WhisperX, with CUDA acceleration for NVIDIA RTX 4070.

Updated to support the latest version of whisperx as of 4/1/2026.

Transcripts are saved as both .srt and .txt files in a transcripts/ folder next to the script.

Requirements

• Windows 11 (AMD64)
• NVIDIA RTX 4070 (or compatible GPU)
• Git Bash
• Python 3.12.0
• CUDA 12.8.0
• cuDNN 9.x (I tested with 9.20 and 9.15)

Setup

1. Install CUDA Toolkit 12.8.1

Download and run the installer:

wget https://developer.download.nvidia.com/compute/cuda/12.8.1/network_installers/cuda_12.8.1_windows_network.exe

After installing, set the CUDA_PATH environment variable if it wasn't set automatically:

• Value: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8

2. Install cuDNN 9.20

Download and run the installer:

wget https://developer.download.nvidia.com/compute/cudnn/9.20.0/local_installers/cudnn_9.20.0_windows_x86_64.exe

Add the cuDNN bin folder to your system PATH:

• C:\Program Files\NVIDIA\CUDNN\v9.20\bin\12.9

3. Install Python 3.12.0

Download from python.org or use the Python Launcher:

py install 3.12.0

4. Run the install script

Open Git Bash in the project directory and run:

source install_4070_complete.bash

This script will:

• Create and activate a Python 3.12.0 virtual environment at .venv
• Install PyTorch and torchaudio 2.8.0cu128
• Verify CUDA 12.8 and cuDNN 9.x are detected correctly — if not, it will download the installers for you and exit
• Install WhisperX with all its dependencies.

Note: If CUDA/cuDNN are not installed when you run the script, it will download the installers to the project directory and exit. Install them, then re-run the script.

Running the app

With the virtual environment activated:

source .venv/Scripts/activate
python whisperx-gui.py

Usage

1. Click Add Files to add audio or video files (.mp4, .mkv, .mov, .wmv, .avi, .flv, .mp3, .wav, .aac, .flac, .ogg)
2. Select a Model — large-v2 is the most accurate, tiny is the fastest
3. Select the Language of the audio
4. Select a Compute Type — float16 is recommended for CUDA
5. Click Transcribe All

Output files are saved to transcripts/<filename>_<timestamp>/ and the folder opens automatically when transcription completes.