MotionShine01 / Drop
Choose a recording
A sound file or a video. Pick the language that is spoken in it.
Free tool
Transcribe audio to text.
Turn a voice memo, an interview, a podcast or a meeting into text. Copy it, keep the timestamps, or save SRT and VTT subtitles. Whisper, an open AI model, runs in this tab, so your recording is never uploaded. No sign-up, no length limit.
This runs in your browser. Your recordings are never uploaded.
How it works
Three steps, nothing to install.
02 / Transcribe
Let the AI listen
The first time, the page downloads the AI model, about 95 MB. Then your computer writes the text, 30 seconds at a time.
03 / Save
Copy or download
Fix any word in the box, then copy it or save plain text, SRT or VTT.
Questions
Good to know.
Is my recording uploaded?
No. The AI model is downloaded to your browser and runs there, so the recording never leaves your computer. There is no account, no minute limit and no watermark.
Which AI model is it?
Whisper base, the speech recognition model by OpenAI, released openly. It knows 99 languages. This is one of the small versions of Whisper, chosen so it downloads quickly and runs in a browser; the larger versions are more accurate but too big for a web page.
How accurate is it?
Clear speech in English and other widely spoken languages usually comes out well, with a few wrong words to fix. Strong accents, crosstalk, music under the voice and less common languages make more mistakes. Clean up a noisy recording first with the background noise remover.
How long does it take?
It depends on your computer. In our test on a Mac it wrote about 3 seconds of speech every second, on the graphics card and on the processor alike. The page shows the time left while it works.
Does it know who is speaking?
No. It writes what is said and when, but it does not tell speakers apart.
What are SRT and VTT?
Subtitle files: each line of text with the time it should show. YouTube, Vimeo, most video editors and players read SRT; web pages use VTT.
Can it translate the text?
Yes. After it is done, pick a language under Translate to. Each line is translated and keeps its time. English to 13 languages and 18 languages to English; the translation model for each pair downloads once, about 115 MB.