Privacy First
Uploaded files are automatically and permanently deleted within two hours.
Automatic deletion
Voice to Text
Dictate a voice note, a reminder, a message for a colleague or a short memo straight into the browser, or upload a recording, and receive the words as a plain-text file (.txt, UTF-8). Transcription runs on Whisper, the open-source speech recognition model, on our own servers – in roughly 100 languages, detected automatically.
-
Record: Speak into the browser recorder – up to 30 minutes per take – or upload MP3, WAV, M4A, AAC, OGG, FLAC or WebM.
-
Transcribe: Whisper writes the words with punctuation; optionally mark speaker changes for interviews and meetings. Output: a plain-text file (.txt, UTF-8).
-
Private: Nothing is uploaded until you confirm. Audio and transcript are deleted automatically after processing, at the latest within a few hours.
Transcription
How voice to Text works
The page combines an in-browser recorder with a GPU transcription backend. Recording uses the MediaRecorder interface built into Chrome, Firefox, Edge and Safari; recognition is done by Whisper, the open-source model published by OpenAI, running on our own hardware rather than on a third-party API.
How to use Converter App
1
Record or upload
Press "Record with your microphone", allow access once, and talk. A timer and level meter confirm that audio is coming in. Or upload an existing recording – voice memos from any phone are fine.
2
Check and start
Listen to the take in the dialog. Switch on "Detect multiple speakers" for interviews and meetings (it marks who is talking but takes noticeably longer). Then start the transcription.
3
Download the transcript
Progress is shown while the model works; a few minutes of speech usually take well under a minute. Download a plain-text file (.txt, UTF-8) from the result page.
Key features
Whisper under the hood
Whisper is trained on 680,000 hours of multilingual audio and handles accents, background noise and technical vocabulary far better than the dictation engines of a few years ago. The language is detected automatically.
Speaker detection on request
For conversations, the optional speaker detection labels changes of speaker in the transcript. It roughly doubles the processing time, which is why it is off by default.
Text output that fits your workflow
The transcript is saved as a plain UTF-8 text file with normal punctuation and paragraph breaks, so it opens in any editor, word processor or note app and can be pasted anywhere.
Free and private
No account, no watermark, no trial counter. The recording never leaves your device until you confirm, and audio and transcript are removed automatically after processing.
Experience & privacy
Built for reliable transcription workflows.
Developed by engineers with 10+ years of experience in large-scale infrastructure, data systems, and scientific computing. Designed for real-world audio workflows where privacy, dependable processing, and practical usability matter.
Trusted by Users
Rated 5 stars on Trustpilot for speed, reliability, and ease of use.
User trust
Academic Use
Referenced in published research and used for interview transcription and qualitative data analysis.
Research use
How voice becomes Text on this page
Your browser records the microphone as Opus audio (Chrome, Firefox, Edge) or AAC (Safari) and keeps it in memory. When you confirm, the file is uploaded over HTTPS to one of our transcription servers.
There, ffmpeg decodes the audio and Whisper – an open-source speech recognition model – turns it into text in one pass, including punctuation and sentence breaks. If speaker detection is on, a diarisation model first splits the audio by speaker and the transcript is annotated accordingly.
The transcript is saved as a plain UTF-8 text file with normal punctuation and paragraph breaks, so it opens in any editor, word processor or note app and can be pasted anywhere. The result page lets you download it; the audio and the transcript are deleted automatically afterwards.
Recording tips that matter more than any setting
Speech recognition is only as good as the audio. Three things make the biggest difference: distance, noise and overlap. Keep the microphone 10–20 cm from your mouth; a phone held close beats a laptop across the table. Switch off fans, close the window, and avoid rooms with strong echo.
Speak at a normal pace and finish sentences – the model uses context, so half sentences and filler words produce worse text than slightly slower, complete speech. For interviews, ask people not to talk over each other; overlapping speech is the hardest case for any recogniser.
Listen to the preview before you start. If the level meter barely moves or is constantly at full height, re-record rather than hoping the transcript will come out right.
Typical uses for voice to Text
This tool is used by people who prefer to talk rather than type, or who need a voice message outside a chat app: drafting an e-mail or article by talking, capturing meeting minutes, turning a lecture or a voice memo into searchable notes, transcribing interview answers for research, and getting a readable version of a voice message that cannot be played right now.
Because Whisper detects the language automatically, the same page works for a German dictation, a Spanish voice note or an English lecture without any setting. The transcript is not translated – it is written in the language that was spoken.
For audio that already exists as a file – podcasts, lectures, calls – the audio-to-text and MP3-to-text converters on this site run the same engine with the same options.
What this tool does not do
It does not translate: transcripts are in the spoken language. It does not produce subtitle timings – for SRT or VTT files use the subtitle converters on this site. It does not identify speakers by name, only by change of speaker when speaker detection is on. And it cannot recover words that are inaudible in the recording.
Proper names, product names and rare abbreviations are the most frequent errors; a quick read-through of the result with those in mind is usually all the editing a clean recording needs.
FAQ
Frequently Asked Questions
How accurate is the transcript?
With a clear recording and one speaker, Whisper typically gets well over 90 % of the words right in major languages, including punctuation. Accuracy drops with strong background noise, several people talking over each other, very quiet audio or heavy dialect – the recording tips below help more than any setting.
Which languages are supported?
Whisper recognises roughly 100 languages and picks the language automatically from the first seconds of audio, so there is nothing to set. Mixed-language recordings are transcribed in the dominant language. The transcript is written in the language spoken; it is not translated.
Is the audio uploaded while I speak?
No. The recorder keeps the audio in your browser until you press the button in the dialog. If you cancel or close the tab, nothing has been sent. Uploaded files and transcripts are deleted automatically after processing.
How long can the recording be?
The in-browser recorder stops after 30 minutes; uploaded files can be longer. Processing time grows with the length of the audio, and speaker detection adds roughly the same amount again.
What does "Detect multiple speakers" do?
It runs a speaker-diarisation step before transcription and inserts a marker whenever the speaker changes. Use it for interviews, meetings and podcasts; leave it off for dictation and voice notes, where it only adds waiting time.
Can I edit the text afterwards?
Yes – the result is a plain .txt file. Open it in any editor, fix names or technical terms the model misheard, and paste it where you need it. Proper nouns and rare abbreviations are the most common things to check.