Transcribe audio / video using faster whisper, aim to make it very simple and straight forward. No fancy unnecessary functionalities.
Internet needed for the first run to download the AI model. No internet required after the first run, no data collected.
Built with faster-whisper 'medium' model and PySide6.
Windows exe file now availble here
- Transcribes audio/video files (
.mp3,.wav,.m4a,.flac,.ogg,.mp4,.mkv) into a.txtfile. - Optional timestamps per segment.
- Saves output to an
Output/folder next to the script.
- ~ 1.6 GB of disc space (126 MB for the exe file, 1.42 GB for the Whisper model)
- Internet connection on first run (to download the model)
- At least 1 GB of available RAM when running, or else the transcription process could be slow.
- CPU only — the app runs on CPU only, I choose to do so because most people don't have GPU + CUDA installed, I aim to provide a simple tool for the majority.
- Output is append-only — if you transcribe the same file twice with the same output filename, the result is appended to the existing file rather than overwritten.
- Language auto-detected — the model detects the spoken language automatically. Accuracy may vary, this website has some data about it Whisper-accuracy. English audio seems to have the highest accuracy.
- Just to re-emphasize that your computer will download the model when you run it for the FIRST time - Downloading takes minutes (see stats below), but loading the model will takes only a few seconds.
- For windows - Delete the exe file, and the model that's located in "C:\Users[your name].cache\huggingface\hub"
-
Downloading the model (only for the first run)
- Internet download speed: 50Mbps - It took me 4 minutes 40 seconds
-
Transcribe
- Windows 11, 13th Gen Intel(R) Core(TM) i9-13900H (2.60 GHz) 32.0 GB RAM
- English, 42 seconds, mp3 file — 15.54 seconds, 963.15 MB RAM
- English, 5 minutes, mp3 file — 124.31 seconds, 968.22 MB RAM
- Windows 11, 13th Gen Intel(R) Core(TM) i9-13900H (2.60 GHz) 32.0 GB RAM
