Free
Text Summarizer
Pull the key sentences out of articles, documents and long text for a short, clear summary.
Speech to Text Transcriber
Turn audio, video and voice recordings into editable text with the Whisper AI model running on your own device. Save it as TXT, or as SRT and VTT subtitles with timestamps.
Add audio
Drop an audio or video file
or choose a file
MP3 · WAV · M4A · OGG · FLAC · WEBM · MP4 · MOV
0:00
Your browser asks for permission to use the microphone. The recording stays on your device, and when you stop, it is ready to transcribe.
Not private: your voice is sent online. Live dictation uses your browser's built-in speech service, not the on-device Whisper model. In Chrome and Edge your audio is sent to Google's or Microsoft's servers to be recognised. For private transcription, use File or Record.
Your browser doesn't offer live dictation. Try Chrome, Edge or Safari, or use File or Record, which work in every modern browser.
Starting…
With File and Record, the Whisper AI model runs on your device and your audio is never uploaded. Only the model downloads, once, from Hugging Face, and your browser keeps it for next time.
Transcript
Add a file or make a recording, then press Transcribe. Your editable transcript appears here.
–Length
0Words
0Lines
–Made with
Hearing…
Click any line to correct it. Click a time to play the audio from there.
About the tool
How the Speech to Text Transcriber works
The Speech to Text Transcriber turns spoken audio into text you can edit, copy and save. Drop in an audio file such as an MP3, WAV or M4A voice memo, or a video such as an MP4 or WEBM screen recording, or record straight from your microphone. When the transcript is ready, you can correct any word, click a timestamp to hear that part again, then copy the text or download it as a plain TXT file or as SRT and VTT subtitle files for YouTube, video editors and media players.
The transcription is done by Whisper, an open-source AI speech recognition model from OpenAI, running inside your browser with the Transformers.js library. The first time you use it, your browser downloads the model you choose (about 40 MB for Tiny, 80 MB for Base or 250 MB for Small) and keeps it for next time. Whisper understands 99 languages: it can detect the language for you, or you can pick one, and it can also translate what is said into English. If your browser supports WebGPU, the tool can use your graphics card to run faster.
Because the model runs on your own device, your recordings are never uploaded to a server, which makes it a good fit for interviews, meetings, lectures and private notes. Speed depends on your device: on a typical laptop the Tiny and Base models usually transcribe faster than real time, while phones and older computers are slower. Accuracy is best with clear speech and little background noise, so it is worth reading the transcript through before you use it. The page also offers your browser's own live dictation for quick notes, with a clear warning, because browsers like Chrome send that audio to an online speech service.
Frequently asked questions
Is my audio uploaded to a server?
No. Your file or recording is decoded and transcribed on your own device by the Whisper model running in your browser. Only the model and the code that runs it are downloaded, from Hugging Face and jsDelivr. The optional Live dictation mode is different: it uses your browser's built-in speech service, which in Chrome and Edge sends audio online, and the page warns you before it starts.
How big is the download and does it happen every time?
Tiny is about 40 MB, Base about 80 MB and Small about 250 MB, plus about 7 MB for the speech engine. With the graphics card option the files are larger. Your browser stores them after the first time, so next time the model loads from your device in a few seconds, as long as you don't clear your browser data.
Which model size should I choose?
Start with Base: it is accurate for clear speech in common languages and fast on most laptops. Tiny is quicker and good for short notes or slower devices. Small makes fewer mistakes with accents, background noise and less common languages, but it is a much bigger download and slower, so it works best on a computer with WebGPU.
How long can my recording be?
There is no fixed limit, but long files need memory and time. Recordings of an hour or so usually work on a laptop, though they take a while with the larger models. On phones, keep recordings under about 20 minutes. Whisper works through the audio in 30-second pieces, and the progress bar shows how far it has got.
Can I make subtitles for a video?
Yes. Add the video file and the tool takes the sound from it. After transcribing, correct any mistakes in the transcript, then download the SRT or VTT file. SRT works with YouTube, Premiere Pro, DaVinci Resolve and VLC, and VTT is the standard format for subtitles on web pages.
Why is the transcript not perfect?
Whisper is very good but not flawless. Mumbled speech, several people talking at once, music, strong accents and names it has never heard can cause mistakes, and in long silences it can sometimes repeat a line. Try a larger model or choose the language yourself instead of auto-detect, and always check the text before you share it.
More free AI tools
- Productivity
- WritingFree
Case Converter & Word Counter
Change text to UPPERCASE, lowercase, Sentence or Capitalized Case and count words, characters and more.
- ImageFree
Image Converter
Convert images between JPG, PNG, WEBP, AVIF, GIF, ICO, PDF and more, and resize, crop or rotate them in bulk.