About RecToText
Last updated: November 16, 2026
Who builds RecToText
RecToText is an independent project built and run by one person, Fuqing Zhang. There is no company or team behind it: I write the code, answer support email, and decide what gets built next.
I built it to be the quickest way to turn a recording into text: open the page, drop in a file, get a transcript. No account, no install, no upsell before you see a result. If something doesn't work or a transcript looks wrong, you can reach me directly at zfuqing66@gmail.com.
How transcription works
- Your browser uploads the file over HTTPS to private temporary storage (Vercel Blob) that only the RecToText server can read.
- The server converts the audio to a compact mono MP3 (video files have their audio track extracted) with FFmpeg. The uploaded file and the converted copy are deleted as soon as the transcript is ready.
- The audio is sent to Groq, which runs Whisper large-v3, OpenAI's open-source speech recognition model, and returns the text with segment timestamps.
- The transcript is sent back to your browser, where you can copy it or download it as TXT, SRT, DOCX, or PDF.
How accurate is it?
Whisper is one of the most widely used open speech recognition models, but no automatic transcription is perfect. From what I see in practice:
- Works best on clear speech from one speaker at a time, recorded close to the microphone: voice memos, lectures, interviews, podcasts.
- Gets worse with loud background noise or music, several people talking over each other, phone-quality audio, and heavy accents mixed with rare names or jargon.
- Does not label who is speaking (no speaker diarization) yet.
The language is detected automatically, or you can pick one of 13 languages before uploading, which usually helps with short or mixed-language clips. Timestamps are included so you can jump to the original audio and check any passage that matters. I haven't published formal accuracy benchmarks yet; when I do, they'll be linked here with the test files and method.
What happens to your files and data
- Audio: RecToText does not keep your audio. Your browser uploads it to private temporary storage (Vercel Blob, readable only by our server), and it is deleted as soon as transcription finishes. If something fails mid-way, leftover files are deleted automatically within one day.
- Speech recognition provider: to produce the transcript, the audio is sent to Groq and is processed under Groq's privacy policy. I don't control how long a provider may keep API inputs for its own abuse-monitoring purposes, so please don't upload recordings you are not allowed to share with a third-party service.
- Transcripts: not stored on the server. Your recent transcripts are saved only in your own browser (local storage) so you can find them again under Your Transcripts; clearing your browser data removes them.
- From AI assistants: RecToText can also be used from AI assistants such as ChatGPT. The assistant sends a temporary link to your file; it is downloaded, transcribed and deleted the same way, and the transcript goes back to the assistant.
- Free usage limit: an anonymous counter tied to your IP address enforces 3 free transcriptions per day and expires automatically.
- Accounts and payments (optional): if you sign in, your email address is stored to send login links and to know which plan you're on. Payments are processed by Creem, the merchant of record; I never see your card details.
Full details are in the Privacy Policy.
Limits and pricing
The free tier allows 3 transcriptions per day, up to 30 minutes or 200MB per file, with no sign-up. Paid plans remove the daily transcription cap and raise the AI chat and translation limits (file length and size limits are the same on every plan); they pay for the speech recognition API and hosting. See Pricing.
Contact
Questions, bug reports, or a transcript that came out wrong: email zfuqing66@gmail.com. I read every message myself.