Turn recordings into text with OpenAI's Whisper speech recognition model running right in your browser. Drop in an MP3, WAV, M4A or video file, pick the language and get an editable transcript with timestamps, ready to copy or save as TXT, SRT, VTT or DOCX. Your audio never leaves your device.
Last updated:
Preparing…
Transcript
Add an audio or video file. Recordings, voice memos, interviews, podcasts, lectures.
Pick a model and language. The first run downloads the Whisper model once; after that it is cached by your browser.
Press Transcribe audio. Text appears segment by segment while the audio is processed on your device.
Fix any words by clicking into the text, then copy it or download TXT, SRT, VTT or DOCX.
Click a timestamp to play from there. Edits you make in the text are used in every download.
Advertisement
Advertisement
How to use the Audio to Text
Add your recording: drop an audio or video file on the box or click Choose audio file.
Choose a model: Tiny is quickest, Base is more accurate for accents and noisy audio.
Set the spoken language or leave Auto-detect, and choose Transcribe or Translate to English.
Press Transcribe audio. The first run downloads the model once; text then appears part by part.
Correct any words in the transcript, then Copy text or download TXT, SRT, VTT or DOCX.
How in-browser transcription works
This tool runs Whisper, the open speech recognition model released by OpenAI in 2022 and trained on 680,000 hours of labeled audio. A version converted for the web runs inside your browser with WebAssembly, so the recording is processed on your own computer or phone:
Your file is decoded by the browser and converted to 16 kHz mono, the format Whisper expects.
The audio is cut into windows of up to 30 seconds. Each cut is placed at the quietest moment near the end of the window, so words are not split in half, and silent windows are skipped.
Each window is turned into a log-Mel spectrogram and decoded into text with timestamps, which are shifted back to the window's position in the recording.
Choosing a model
Model
Parameters
Download (once)
Best for
Whisper Tiny
39 million
about 40 MB
Clear speech, quick drafts, phones and older laptops
Larger Whisper models (small, medium, large) are more accurate still but are hundreds of megabytes to several gigabytes, which is too heavy for most browsers.
Tips for a cleaner transcript
Pick the language instead of Auto-detect when you know it. Detection uses the first seconds of each part, so music intros or a different-language greeting can confuse it.
Trim long silences and music with the Audio Cutter; Whisper sometimes invents short phrases like "Thank you." in silent or music-only parts.
Use Base for noisy or accented audio. It takes longer but usually fixes names, numbers and short words that Tiny misses.
Dictating live? For speaking into the microphone instead of transcribing a file, use Speech to Text.
Frequently asked questions
How do I convert an audio file to text for free?
Open this page, add your MP3, WAV or M4A file and press Transcribe audio. The Whisper speech recognition model runs in your browser, so there is no account, no upload and no limit on the number of files.
Is my audio uploaded to a server?
No. The audio is decoded and transcribed on your own device. The only download is the Whisper model itself (about 40 MB for Tiny or 80 MB for Base), which your browser caches after the first use.
How accurate is the transcription?
Whisper is accurate on clear speech in common languages. Background noise, music, crosstalk, strong accents and technical names reduce accuracy, and the small Tiny model makes more mistakes than Base. Always proofread important transcripts; you can edit the text directly before exporting.
Which languages are supported?
Whisper was trained on 99 languages. This tool lists 40+ common ones, including English, Spanish, French, German, Portuguese, Hindi, Arabic, Chinese and Japanese, plus Auto-detect. Accuracy is highest for widely spoken languages.
How long does it take?
It depends on your device and the model. On a recent laptop the Tiny model often transcribes faster than real time; Base is roughly two to three times slower. Phones and older computers are slower. The progress bar shows an estimate of the time left.
Can it identify different speakers?
No. Whisper writes what is said but does not label who said it. For interviews, use the timestamps to find each answer and add speaker names while editing.
What is the difference between TXT, SRT and DOCX?
TXT is plain text, with optional timestamps. SRT and VTT are subtitle files with start and end times for each segment, for video players and editors. DOCX opens in Microsoft Word, Google Docs and LibreOffice.
Are all the tools on FreeToolCabin tested for quality and accuracy?
Yes! Every tool on our site undergoes automated Q/A testing in real browser environments to ensure high precision, correct output, and zero errors. If you happen to spot a bug or have a suggestion, please contact us at [email protected] and our team will review it right away.