Keep the best available source
Start with the original recording when possible. Repeatedly compressing or converting a file cannot recover words that are already obscured. Listen to a short passage: are the voices audible, and do several people speak at once? Background music, clipping and distant microphones make review more important. There is no need to upload a video elsewhere just to extract its audio. This tool reads a supported audio track locally.
Choose the language deliberately
If you know the spoken language, select it. Auto Detect examines the opening audio and keeps one language for the recording. A long musical introduction or a recording that changes language can make that choice unreliable. English and French use the general multilingual model. Arabic uses a dedicated, larger local model. Selecting a website language changes the interface, not the language spoken in your recording.
Allow for the first model download
The first run needs the required model files. Arabic weights total about 1.33 GB, so allow time and bandwidth. Once loaded, a model can be reused in the same page. That does not guarantee that a fresh page will work offline. Keep the tab open while processing, avoid other memory-heavy tasks and save your result before closing or starting another file.
Review meaning before grammar
Read along while listening to passages that matter. Check names, dates, numbers, negatives and specialist vocabulary first: one missing “not” can reverse a statement. A confident-looking transcript is not proof of accuracy. English and French correction can improve writing, but it cannot verify what was said. Review suggestions individually; do not turn an intentionally informal quotation into a different statement.
Choose the right export
Copy Text and TXT use the selected readable version, including accepted changes in the Working version. SRT, VTT and JSON preserve the original timed segments. Editing the document therefore does not fix subtitle wording. If you need publication-ready captions, export them and review text, timing and line breaks in a dedicated subtitle editor. Keep your original recording until that review is complete.