Privacy Policy

Version of 6 September 2026

1. Who processes

Isa Dagestani. Contact: @isa_dagestani

2. What data

• Telegram ID and username — to tell users apart and account for minutes
• Voice recordings — while the translation runs
• Recognised text and translation — while the translation runs
• Settings: chosen languages, state of the "my voice" feature
• Job log: time, length, language pair, cost, outcome. Conversation content is not logged
• Payment data: order number, amount, currency, payment system

3. Why

Delivering the translation service. Accounting for paid minutes and resolving payment disputes. Finding failures and monitoring quality.

4. Who receives it

• ElevenLabs (USA) — speech recognition and synthesis: the recording and the translated text
• Anthropic (USA) — translation: the recognised text
• Payment systems — amount, currency, order number

No one else. Never used for advertising or profiling.

5. Quality review

The owner reviews translations to find mistakes: the original recording, the recognised text and the result. Nothing is shared with anyone else and nothing is used for advertising.

Before the first translation the bot states this plainly. The review cannot be switched off: it covers every translation the service makes.

If this does not suit you, do not send recordings; data already sent is removed with /delete_me.

6. Retention

• Recording and translated audio — files are deleted right after delivery. Unfinished jobs after 6 hours. The job log keeps a reference to the recording in Telegram's own storage: it is what lets the bot finish an interrupted job
• Voice samples — while the voice model is being built and refined, and no longer than 7 days from the last recording sent. Switching "my voice" off or deleting your data erases them at once
• Voice model — until the feature is switched off or data is deleted
• Job log and settings — until the user deletes their data
• Payment data — as required by accounting rules

7. The voice model

The "my voice" feature builds a voice model from the user's recordings. This is biometric data. The feature is on by default: the bot says so in the chat the moment it starts collecting recordings, with a switch-off button in the same message. It can be turned off at any time — /voice, the bot settings or the mini app.

Samples are kept while the model is being built and refined: a voice taken from ninety seconds of speech sounds coarser than one built from three minutes. Once the final model is built, the feature is switched off or the data is deleted, they are erased. The model is used only for that user's translations and is never shared.

To delete the model use /voicereset or /delete_me. Deletion covers the copy held by the synthesis contractor.

8. Rights

• Find out what is stored — ask @isa_dagestani
• Delete data — /delete_me. Only payment records remain
• See what is reviewed and why — /privacy

9. Security

Data travels over secure connections. Only the owner has access to the database. Backups are made twice a day and kept for 30 days.

10. Changes

The current version is at https://voicetranslator.halal.tools/legal/privacy and via /privacy in the bot.