Mongolian Dictation Privacy: How Duudlaga Flow Protects Your Voice Data
Duudlaga Flow never trains AI on your audio, encrypts your transcripts at rest, and permanently deletes your data when you ask. Learn how we protect Mongolian dictation privacy.
Your voice is not training data
When you dictate in Mongolian with Duudlaga Flow, your voice recordings and transcribed text are not used to train AI models by default. The setting that would allow it ships switched off, and it only ever covers recordings made after you turn it on yourself โ see our announcement on the Mongolian speech model.
This is not a marketing line. It is built into how the service works and into the contracts we have with the companies that process your data.
Encrypted so even we cannot casually read it
Every transcription, note, and uploaded file is encrypted at rest with a key unique to your account.
In simple terms: if someone opened the database, they would see ciphertext, not your words. Your content is decrypted only when it is sent back to you.
This applies to:
- Live dictation transcripts
- Notes you create inside the app
- File transcription results
- Uploaded audio files (encrypted with envelope encryption)
Delete means deleted
When you clear your data or delete your account, the actual content is permanently erased.
There is no hidden trash can. There is no "recover within 30 days" backup. There is no shadow copy.
We keep only the minimum billing and quota records required to maintain accurate account history. The words themselves are gone.
Where your data actually goes
Only two outside services can touch your data, and only what they need:
| Provider | Receives | Purpose |
|---|---|---|
| Your voice recording while you dictate | Converts speech to text | |
| OpenAI | The transcribed text only | Polishes, formats, translates, or runs AI text actions you request |
Important details:
- Your audio never goes to OpenAI. OpenAI receives text only.
- For live dictation, your audio goes to Google only.
- Google and OpenAI are both contractually barred from using your data to train their models.
- Two optional features use other providers on your explicit request: speaker separation (Speechmatics) and text-to-speech (ElevenLabs or Google). Speechmatics' own terms do permit it to use resulting transcripts to improve its service, so that feature is off by default and chosen per file. Section 3 of our Privacy Policy sets out each provider in full.
On-premise option for businesses
Companies that want dictation data to stay inside their office can use the Duudlaga Flow on-premise deployment. The entire platform runs locally except for the speech-to-text call. Polish, formatting, and file storage stay on your own network.
Why this matters for Mongolian users
Mongolian speakers use dictation for work notes, meeting transcripts, creative drafts, and deeply personal thoughts. A tool that handles that data must be built on trust before it is built on speed or features.
Duudlaga Flow is designed so that your voice, your thoughts, and your words stay yours.