WhisperX Transcriber: a free, portable Windows GUI built on faster-whisper

[ibrahimqureshae]

Hi everyone 👋

I'll start with a genuine thank you to everyone who works on faster-whisper. It's the engine doing the actual heavy lifting in my project, and the fact that it runs quickly on plain CPUs and gets so much out of a GPU is what made the whole thing feel practical to build in the first place.

Quick context on the stack: I built a desktop app for people who'd never open a terminal. It uses WhisperX for the word level alignment, and WhisperX runs on faster-whisper underneath, so this is really sitting right on top of your work.

🔗 https://github.com/ibrahimqureshae/whisperx-transcriber

What it is

A small portable Windows app, around 19 MB. No installer, no admin rights, and you don't need to know any Python. You unzip it, double click, and a one time wizard sets up Python and the engine for you if you don't already have them. After that it runs completely offline.

It pulls the Systran/faster-whisper models straight from the hub and caches them next to the app, so the first run grabs the model once and everything after that works with no internet at all.

What it does, with a few faster-whisper specific notes

• Model picker from tiny up to large-v3, with the cache kept so you never re-download
• Sensible compute types: float16 on CUDA, and it falls back to int8 on CPU automatically instead of erroring out
• Auto-detects an NVIDIA GPU and installs the matching CUDA build, or runs on CPU if there isn't one
• Word level timestamps exported to SRT, VTT, TXT, TSV and JSON, plus a per word JSON with confidence scores
• 20 languages in the dropdown, all 99 through the bundled CLI, and right to left scripts like Arabic, Urdu, Persian and Pashto work properly

How it's built, if you're curious

The exe ships with none of the ML packages inside it, just a tiny GUI shell. On first launch it pip installs the stack (which pulls in faster-whisper and CTranslate2) into a local runtime venv next to the app. That keeps the download small and lets the engine update on its own without me re-shipping everything. All the inference logic lives in a small GUI free module that the app and a CLI both share.

It's still early and I'd love feedback

I'm a working student doing this in my spare time, so it's MIT licensed and I'm pretty open about where it goes. Bug reports and feature ideas are very welcome, here or in the issues. Next on my list: a macOS build, drag and drop, batch folders, speaker diarization, and a translate to English mode.

If it turns out useful to you, a star helps other people find it. Thanks for reading, and thanks again to everyone behind faster-whisper. 🙏