How to transcribe audio or video to text
- Choose or drop an audio or video file.
- Pick the quality and the spoken language (or leave it on automatic), then click Transcribe.
- Fix any mistakes, then download the text, an SRT or VTT subtitle file, or add the captions to your video.
What you can do with it
- Transcribe interviews, lectures, meetings, podcasts and voice notes into text you can copy and edit.
- Make subtitles for YouTube, TikTok, Reels and other players. SRT works almost everywhere; VTT is the format for web video players.
- Add captions to a video so they are always visible, which is what most people want for social posts watched with the sound off.
- Translate to English: tick the box and speech in another language is written out in English.
Your recording is never uploaded
Most transcription sites upload your audio to their servers and charge by the minute. Here the open-source Whisper speech model is downloaded to your browser and your own computer or phone does the listening. Nothing you say leaves your device, so it is safe for private meetings, interviews and personal notes, and there is no minute limit, account or watermark.
Fast or more accurate?
You choose between two sizes of the model. Both are downloaded only the first time you use them.
| Quality | Download | Speed on a typical computer | Best for |
|---|---|---|---|
| Fast | 42 MB | About a third of the length of the recording | Clear speech, quick drafts, phones |
| More accurate | 76 MB | About the length of the recording | Accents, background noise, other languages |
So a 10-minute recording takes roughly 3 to 10 minutes on a laptop. Phones are several times slower and may run out of memory on long files, so on a phone use Fast and keep recordings to a few minutes. Keep the tab open and in front while it works. You can press Cancel at any time and keep what has been transcribed so far.
How accurate is it?
With one person speaking clearly into a decent microphone, the text is usually very close and needs only light corrections. It is honest to say what makes it worse:
- Names, brands and technical terms are often misspelled. Check them.
- Background music, noise, crosstalk and strong accents cause mistakes, especially with the Fast option.
- Speakers are not labelled. The text does not say who said what.
- Silence and music can occasionally produce a stray phrase that nobody said. Delete it.
- Subtitle timings are approximate, usually within a second. They suit captions well, but they are not frame-accurate.
These compact models are far smaller than the ones behind paid transcription services, so expect more errors than those on difficult audio. English is the strongest language; Spanish, French, German, Italian, Portuguese and other widely spoken languages work well, and less common ones are weaker.
Need to cut the recording first? Use the audio trimmer or the video trimmer. To pull the sound out of a video, use Video to MP3.
Frequently asked questions
Is this speech to text tool free?
Yes. It is free with no sign-up, no minute limit and no watermark.
Is my audio or video uploaded?
No. The speech model runs in your browser and your recording never leaves your device.
How do I make an SRT subtitle file?
Transcribe your video, choose SRT subtitles above the text box, fix any mistakes and click Download SRT. Upload that file alongside your video on YouTube and most other platforms.
How do I add captions to a video?
Transcribe the video, then click Add Captions to Video. A new MP4 is made with the subtitles drawn onto the picture. It is made on your device, so allow a few minutes for longer videos.
How long can the recording be?
There is no fixed limit, but the work is done by your device. Up to an hour is practical on a computer. On a phone, keep it to a few minutes.
Which languages does it understand?
About 100, including English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Arabic, Hindi, Chinese, Japanese and Korean. Accuracy is best for widely spoken languages. It can also translate speech into English.
Why is it slower than other sites?
Other sites run huge models on server farms and upload your recording to do it. This tool keeps your recording private by running a compact model on your own device, which takes longer.