Voice Typing and Privacy — What Actually Happens to Your Audio
· privacy, security, audio, transparency
People are right to ask what happens to their voice. Dictation involves sending audio somewhere, and "somewhere" deserves a clear answer rather than a marketing line. Here is how browser-based voice typing works and what to check in any tool you use.
The path your audio takes
- Your browser asks for microphone permission. Nothing is captured before you grant it and press record.
- Audio is captured in short segments while you speak.
- Each segment is sent over an encrypted connection to a transcription model.
- Text comes back and is displayed.
- In Baani AI the audio segment is discarded once transcription completes — it is not kept as a recording.
What is stored
- Transcripts you keep are stored so history works. History lives with your account, and you can clear it at any time.
- Saved Words are stored so corrections persist across devices.
- Audio is not retained after transcription.
Our Privacy policy is the authoritative statement of this.
What to check in any dictation tool
- Does it say plainly whether audio is retained, and for how long?
- Is your content used to train models? If the answer is not stated, assume it might be.
- Can you delete your history, and does deletion actually remove it?
- Is the connection HTTPS? Browsers refuse microphone access otherwise, so any tool that works at all clears this bar.
Dictating sensitive content safely
- Do not dictate passwords, OTPs, card numbers or bank credentials into any tool
- For medical, legal or HR content, dictate anonymised facts and add names by hand afterwards
- Be aware of your physical surroundings — the most common privacy failure in dictation is being overheard, not interception
- On a shared device, clear history when you finish
Microphone permission hygiene
Browsers only allow microphone access to the tab you granted it to, and only while the page is open. You can revoke it at any time from the padlock icon in the address bar. If a page could listen without a visible indicator, your browser would show one — the recording indicator is enforced by the browser, not the site.
The short version
Dictation requires trusting a transcription service with a few seconds of audio at a time. That trust should be based on a stated retention policy you can read, not on assurances. Read ours, and hold every other tool to the same standard.