Steps
Transcribe the recording
Drop the audio or video file (MP3, M4A, WAV, MP4, WebM and similar). Choose the accuracy: Balanced for most recordings, Fast on older phones, Best for accents, noisy rooms and languages other than English. The first run downloads the chosen model once (the size is shown before you start). Pick the spoken language or leave automatic detection on, and press Transcribe; text appears section by section and you can click any line to hear it. Timestamps are switched off so the text reads as plain paragraphs; download the .txt or press Copy text. Keep the tab open while it works.
Turn the transcript into a Word document
Add the transcript text and turn the transcript into a Word (.docx) document. Before sharing it, add speaker names where needed (Whisper does not mark who is speaking) and correct names, figures and technical terms.
Requirements and sources
| What | Value | Source |
|---|---|---|
| Whisper tiny model (columns: size, parameters, English-only model, multilingual model, required VRAM, relative speed) | tiny | 39 M | tiny.en | tiny | ~1 GB | ~10x | openai/whisper - GitHub (project README, Available models and languages) Last verified 2026-10-21 |
| Whisper base model (same columns) | base | 74 M | base.en | base | ~1 GB | ~7x | openai/whisper - GitHub (project README, Available models and languages) Last verified 2026-10-21 |
| Whisper small model (same columns) | small | 244 M | small.en | small | ~2 GB | ~4x | openai/whisper - GitHub (project README, Available models and languages) Last verified 2026-10-21 |
| Whisper - accuracy by language | Whisper's performance varies widely depending on the language. | openai/whisper - GitHub (project README) Last verified 2026-10-21 |
| Whisper - English-only models | The .en models for English-only applications tend to perform better, especially for the tiny.en and base.en models. | openai/whisper - GitHub (project README) Last verified 2026-10-21 |
| Whisper paper - training data | When scaled to 680,000 hours of multilingual and multitask supervision | Robust Speech Recognition via Large-Scale Weak Supervision (Radford et al., arXiv:2212.04356) Last verified 2026-10-21 |
FAQ
Can it transcribe a Zoom, Teams or Google Meet recording?
Yes, if you have the recording file. Meeting apps usually save MP4 video, and some also save an audio-only file such as M4A; either works, and an audio-only file is smaller to process. Download the recording from the meeting app or cloud storage first, then drop it here.
How long does transcription take?
It depends on your device and the accuracy setting. On a modern laptop, Fast and Balanced run faster than real time; phones and the Best setting are slower. Long recordings are processed in two-minute sections, so you can watch the text arrive. Keep the tab open and the device awake.
Which Whisper models does this use, and what is downloaded?
The multilingual tiny, base and small Whisper models (shown as Fast, Balanced and Best), run in the browser with transformers.js. The chosen model is downloaded once from Hugging Face, roughly 45 MB, 95 MB or 200 MB, and then cached by your browser. Your recording is never sent anywhere.
Does it tell me who said what?
No. Whisper writes punctuation and capitals but does not separate speakers. Add speaker names in the Word document; clicking a line in the transcript plays that part of the recording, which makes this quicker.
Can I transcribe a recording in another language, or get it in English?
Pick the spoken language from the list (22 languages plus automatic detection) and use Best for non-English speech, because accuracy varies a lot by language. To get English text from another language, set Output to Translate to English.