Privacy

Zubaan · ਜ਼ੁਬਾਨ — last updated 13 November 2026

Zubaan is a Punjabi learning app. This page says exactly what it stores, why, and who else sees it. If something here is not clear, ask and it will be made clearer.

You can use it without an account

The bundled vocabulary and practice tools work signed out. Online AI features require an account and use its allowances. With no account, your progress is kept on your own device. Zubaan counts onboarding and learning actions such as opening Ask, choosing words, audio or speaking, receiving a library or online answer, saving a word, and starting or completing saved-word practice. Optional “Did this help?” choices are counted too. These counts help us improve the experience. The random journey identifier lasts for one browser tab and is never attached to an account; answers, selected goals, claimed words, names, transcripts, and device details are not sent. An account lets you carry progress between devices, contribute words to the shared dictionary, and use the metered online AI tools.

Ask Zubaan

Library word lookups happen on your device. For an online explanation, your question and the recent conversation are sent to the AI provider through the existing tutor service (or directly if you configured your own key). The Ask conversation is held in this tab's memory and is cleared on reload, New chat, sign-out, or switching accounts. Signing in from a guest session keeps the question you are working on. If you choose to sign in, your current unsent question is temporarily kept in this tab's session storage so it can return after sign-in; it is restored once and expires after 30 minutes. New chat and sign-out clear that draft too. Ask answers are excluded from the on-device AI answer cache. Saved library words are stored as word identifiers on this device; practice results use the ordinary learning-progress system. Questions, pasted lyrics and answers are not sent to Zubaan's analytics or automatically added to the community dictionary. Provider processing is described below.

Song clips, conversations and oral histories

Song clips, family conversations and oral history chapters can be up to 30 minutes / 100 MB. Recording or selecting a clip does not upload it. After you confirm that you can share the recording, pressing Transcribe sends short audio passages through Zubaan's server to the configured transcription provider: Google Gemini, or Sarvam AI for spoken transcription if enabled. Pressing Translate sends the editable transcript as text to Google Gemini. Optional Cross-check sends passages for another model to transcribe, then translates that second transcript separately. Completed checks are reused. Those providers process the material under their service terms; keeping a private draft does not mean the processing happens entirely on your device.

By default Zubaan keeps the recording, transcript and translations in this tab's memory. Clear recording, reload, sign-out or switching away from an account removes that work. Leaving the audio screen stops further processing and keeps completed work while this tab remains open. Download the original audio and transcript before reloading or signing in through a redirect. Save progress + data downloads a private JSON file containing drafts, second-model answers and a fingerprint of the audio. To resume, select the original audio and load that progress file; matching happens on your device without provider requests. A fingerprint matches a recording, not the accuracy of a transcript. Downloads are files you control and are not removed by clearing Zubaan. Zubaan does not save these clips or drafts to its dictionary, cloud storage, analytics or training dataset. Account allowance usage is recorded separately. Processing a clip does not grant permission to publish it or use it for training.

Keep processing if I close this tab is a separate, optional choice for signed-in accounts. When you tick it, the short audio passages are uploaded to a private storage folder that only your account can read, and Zubaan's server transcribes and translates them in the background, saving each finished passage, its transcript and translation in your account. Your edits stay in the tab as before. Zubaan deletes the uploaded audio when you open the finished job or delete the job; a job that is never finished expires after seven days and its audio is deleted the next time you open the audio screen. Transcripts and translations stay in your account until you delete the job or your account. Deleting your account deletes all of it. This storage is not used for the dictionary, analytics or training, and nobody else can read it.

What is stored if you sign in

The app saves a copy of dictionary data for offline use on this device, separately for each account. Sign-out removes that account's downloaded copy; it can be downloaded again. Words you deliberately save in Ask remain for your next sign-in on this device. Deleting your account also removes its Ask saved-word list. These copies support offline learning and do not contain your Ask conversations.

Your email is shown beside feedback to the founder only if you separately allow follow-up contact. That permission is optional, can be changed in Settings, and is not consent to marketing email.

Recordings

Nothing is recorded until you press a record button. What happens next depends on which button it was:

Recording other people. The live translation and mining features are built for sitting with an elder and capturing how they speak. Ask them first. A recording of somebody's voice belongs to them as much as to you, and this app is only worth building if the people in it agreed to be there.

Limits that keep it running

To stop one person or program from using up the shared online AI for everyone, Zubaan's server counts how many requests arrive in each minute and each day, per account and per network connection. For the network count it never stores your IP address: the address is turned into a one-way code using a secret only the server knows, and that code changes every day, so one day's counts cannot be linked to the next. Each count holds only that code or your account ID, a number, and the time window. A count is deleted automatically once its window (a minute or a day) has ended and new requests arrive, and is never used for analytics or anything else. If you reach a limit, the app tells you how long to wait; the bundled library keeps working offline.

Who else your data touches

Nothing is sold, and there is no advertising, no analytics SDK, and no tracking across other apps or websites.

Children

There is a kids mode, meant to be used with a parent. It changes which words appear and how questions are asked; it does not change anything on this page. Zubaan is not aimed at children under 13 as independent users, and no account is needed for a child to learn from it.

Getting rid of it

You can delete your account inside the app: open Settings → Shared dictionary → Delete account. This deletes sign-in, synced progress, your profile, votes, recordings, background audio jobs and their uploaded audio, feedback, and beta survey answers. Learning data already stored on your own device can remain so you do not lose it by accident; Reset everything in Settings removes that copy too. You can also email singhsunpreet300@gmail.com for help.

One exception, stated plainly: words and corrections you contributed to the shared dictionary stay, with your name removed. Other people are learning from them, and pulling them out would damage entries that are no longer only yours. Recordings of a voice are different — say so and those are deleted outright.

Changes

If this page changes in a way that matters, the app will say so rather than quietly updating the date at the top.