Vozport
Privacy Policy
Last updated: November 6, 2026
1. Who is responsible (data controller)
Vozport is provided by 68.216.024 Victor Rosa Lima (MEI), Brazilian company registration (CNPJ) 68.216.024/0001-16, address Rua João Antônio Morato do Amaral, 225, Santa Rita, Piracicaba/SP, CEP 13423-258, Brazil ("we"). Privacy contact (also for the data protection officer role under the LGPD): contato@vrosa.com.br.
2. Data we process and why
- Your speech and typed text (to translate). Each spoken segment (up to 60 seconds; longer speech is split automatically) or typed text (up to 500 characters) is sent to our server and, through OpenRouter, to AI providers for speech recognition, translation and synthetic voice; the translated text goes to Fish Audio only to create the voice: through OpenRouter for the standard voices, and directly from our server for the voices you create (section 4). It is processed in memory only to produce the translation and is discarded after the response. We do not store audio, transcripts or translations on our servers or in our logs, unless you turn on "Help improve the AI" (section 3); the only exception is the voice cache below.
- Voice cache (phrases many people say). To answer faster, when several different people ask for exactly the same short translated phrase, we may keep the synthesized voice of that phrase and reuse it for everyone. A phrase is only kept after at least 3 different people (devices) have produced it, and the stored audio is linked to no one: we do not record who said it, and the phrase is found only through a keyed code (HMAC). Phrases with data that identifies or allows contacting someone are left out: long numbers (any run of 3 or more digits, or 4 or more digits in total — such as phone, document, card or booking numbers), e-mail addresses, links and @handles, as well as long texts. Short numbers (a table, a time) and names only get in if the exact same phrase comes from at least 3 different people, so a detail of a single person does not. Before a phrase is kept, it is also checked by an AI classifier (Jev, from TypeSafe, through OpenRouter), which receives only the text of that phrase — never who said it — and answers whether it could identify or locate a person; if it could, or if the check fails, the phrase is not kept. To count different people we keep, for up to 30 days, keyed counters that do not reveal who they are. So that a thumbs-down on the voice can remove a bad audio, the technical link between a translation and the cached phrases used in it is kept for 30 minutes only. The audio is deleted automatically after 75 days without use, and after at most 180 days in any case.
- Conversations and translation context. Your conversations are saved only on your device: up to 200 conversations (past that, the oldest one is deleted). To keep the translation consistent, at most the last 6 exchanges of the current conversation are sent along with each request and are not stored; the other saved conversations are never sent.
- Conversation title. To make your saved conversations easy to find, the app asks for a short title for each one: it sends the text of up to 12 phrases of that conversation (what was heard and the translation, at most 2,000 characters in total) to our server and, through OpenRouter, to the same AI providers that translate (section 4), only to write the title. The text is processed in memory and is not stored or logged by us; the title is saved only on your device. The title does not count as a phrase.
- Voices you create (Pro). Subscribers (including the free trial) can create synthetic voices for the translations in Settings › Voice › Voices › "Create voice", by describing how the voice should sound (for example, "calm and deep, speaks slowly"). Before the first creation, the app asks for your permission. The description you write (up to 300 characters) and the adjustments you ask for after hearing the previews (such as "deeper") go to our server, which applies its own rules (it refuses e-mail addresses, phone numbers, links and forbidden requests) and then sends them, through OpenRouter, to an AI model from Anthropic (Claude), which checks the request and rewrites it in English as a voice instruction; the check may also use the Jev classifier, from TypeSafe, through OpenRouter. We refuse children's or teenagers' voices, imitations of real people or characters, and sexual, offensive or deceptive content; if any check fails, nothing is created. Once the request is approved, our server sends the rewritten instruction and a fixed sample sentence directly to Fish Audio, which returns previews of the voice. Fish Audio may use the rewritten instruction, the voice model and the texts spoken in that voice under its own terms, which allow it to use this content to develop and train its AI models. It has told us in writing that, by default, it does not train on data received through its API. So that you only hear complete previews, their audio (synthetic, with none of your data) is transcribed automatically by a speech recognition provider from section 4, through OpenRouter; the same applies to a test sentence spoken by the voice when you save it. The description you write is not kept. After the check, our server keeps only the rewritten English version and the short summary the app shows (such as "Got it: female voice, calm"), for at most 24 hours, together with the previews, so you can adjust the voice. The description, the adjustments and the rewritten version never go into our logs or statistics: the logs keep only the result of the check (for example, "refused: child's voice"). The previews stay in our storage for up to 24 hours and are then deleted. The preview you save becomes a private voice model, kept at Fish Audio and linked to your anonymous ID (stored only as a hash); the name you give the voice never reaches our server: it stays on your device and, like your settings, goes into the device's own backup (on Android, the Google backup; on iPhone/iPad, iCloud, when it is on). The voice you create is synthetic, is not made from your voice or anyone else's, and is never shown to other users. When it speaks a translation, the translated text goes directly from our server to Fish Audio (section 4), and its audio does not go into the shared voice cache: it stays only on your device. The voice's samples (one fixed sentence per language) stay in our storage while the voice exists. The model stays at Fish Audio until you delete the voice (Settings › Voice › Voices › "⋯" › Delete voice) or ask us to delete your data (section 8); when you delete it, we ask Fish Audio to delete the model (anything it still keeps after that follows its own terms), we delete the voice's samples and record on our server, and the app deletes from the device the audio saved in that voice. If your subscription ends, the voice is paused and is deleted after 90 days without a subscription. If you report a preview or a voice, we keep the reason, the anonymous ID and the reported synthetic audio for up to 90 days, for review.
- Microphone that reopens once, and settings applied on the device. The microphone only listens while the app shows that it is listening. Besides conversation mode (which listens again after each translation), after the translated voice plays for the other person, the app reopens the microphone once on its own to hear the reply, with the status on screen; if nobody speaks within a few seconds, it closes. You can turn this off in Settings ("Listen for the reply after the voice"). The voice speed (1x, 1.25x or 1.5x) and "Read replies silently" are applied on the device and change nothing in what is sent to the AI providers.
- Anonymous ID. There are no accounts. When you install the app, RevenueCat creates a random anonymous ID. We use it to know whether you have a subscription, to count your usage and to link the usage statistics. It does not contain your name, e-mail or phone number.
- Device identifier (fraud prevention). To count the free phrases per device (so that reinstalling the app does not reset them), the app sends a device identifier: on Android, the Android ID (a number specific to this app on your device); on iPhone/iPad, a random identifier that the app saves in the device's Keychain. Our server keeps only a keyed hash of it, never the identifier itself. It is not used for advertising or statistics and is not shared.
- IP address. Used for security and to apply limits (per-minute limits and a daily limit of free phrases per network, against automated abuse) and, in the usage statistics, to estimate your approximate location (country/city). The limit counters keep only a keyed hash of the address, and our application logs do not record it.
- Technical data. Platform, app version, the languages and mode you chose, response times, errors and the AI models used — never the content.
- Subscription. RevenueCat receives the anonymous ID and the subscription status and history from Apple or Google (product, dates, trial, renewal) to unlock the app. Payment is handled by Apple or Google: we never see your name, e-mail or card.
- Usage statistics (can be turned off). Sent to PostHog, linked to the anonymous ID: app events (e.g. screens viewed, buttons tapped with all text masked, translation completed, response times, errors, purchase steps), device and app information (device type, operating system, app version, language, time zone, screen orientation), the choices made in Settings (such as voice, voice speed and color, never text) and approximate location derived from the IP address. On a sample of app sessions, a session replay records the screens with all text and fields masked and the conversation, full-screen and typing areas blocked. Statistics never contain audio, transcripts, translations or typed text. To stop them (in the app and on our server): Settings › Privacy and data › "Send anonymous usage stats".
- Data saved on your device. Your settings, your conversations, usage counters and the statistics identifier are saved on the device. The app also keeps on your device the audio and text of recent translations, to play them again faster and offline; you can delete them in Settings ("Clear voice cache"; "Delete my training data" also deletes them). "Delete all conversations" (Settings) deletes every conversation and all the voice saved on the device; one conversation can also be deleted on its own (Conversations › "⋯" › Delete). Everything is removed when you uninstall the app (the Keychain identifier on iPhone/iPad may remain on the device). Device backup: the open conversation and your settings (including the names you give the voices you create) go into the device backup, as before. On Android, the list of conversations also goes into the Google backup: for each one, the title (the one written by AI, or the beginning of its first phrase), the date, the languages and the number of phrases; only the full phrases of the older conversations and the saved voice are left out of the Google backup. On iPhone/iPad, everything goes into the device's iCloud backup (when it is on), like the rest of the app's data.
- This website (vozport.com). The website, not the app, counts visits with PostHog, through an address of ours (e.vozport.com) that forwards the data to PostHog. We record the pages viewed, which sections came into view, how far the page was scrolled, and taps on parts of the site (such as the store badges, the language switch, the questions, the demos, the audio buttons and links between pages), together with the page language, the type of system (Android, iOS or computer), technical data from the browser (page address, browser, operating system, language and time zone, screen size, referring page) and, if they are in the link, campaign parameters (utm). Your IP address reaches PostHog, which uses it to estimate your approximate location (country/city). There are no cookies and nothing is saved in your browser (no localStorage): the statistics identifier is random and lasts only while the page is open. There is no session replay and no automatic capture of clicks or text, and nothing you type on the site is sent. If your browser sends Do Not Track or Global Privacy Control, nothing is sent. The website is hosted by Cloudflare, which processes your IP address to deliver the pages.
- Messages to us. If you write to us, we use your e-mail address and message only to answer you.
3. "Help improve the AI" (optional, off by default)
- What is saved: only if you turn it on, after each successful translation we save the audio you sent and a record with the transcript, translation, languages, mode, AI models used, response times, app version, platform, your anonymous ID and the version of your consent.
- Why: to evaluate and train our translation and speech AI models. We do not sell this data.
- Where: in object storage from Tigris Data on Fly.io's infrastructure (United States), accessible only to us.
- For how long: until you delete it, and at most 3 years. We also keep a record of your choice (version, date and platform of the consent) to prove it.
- Bonus: while it is on, your daily usage limit is 10% higher (in the free plan, 28 free phrases instead of 25). Saying no does not change anything else.
- Other people: recordings may include the voice of the person you are talking to. Only turn it on if they agree.
- Withdraw and delete: turn it off at any time in Settings › Privacy and data (nothing new is saved after that). "Delete my training data" withdraws your consent and deletes everything saved (audio and records). The app keeps working normally.
4. Service providers (processors)
- OpenRouter (United States): routes each request to the AI providers.
- AI providers, through OpenRouter: Google (translation and some voices), Fish Audio (voice), xAI (speech recognition and some voices), NVIDIA, Deepgram, OpenAI and Alibaba Cloud/Qwen (speech recognition, depending on the language), Anthropic (checks and rewrites the descriptions of the voices you create, section 2), and the cloud providers that run open models such as Gemma, Whisper and Parakeet (for example DeepInfra, ModelRun and CoreWeave); and TypeSafe (Jev), which checks whether a phrase could identify someone before it enters the voice cache (section 2) and may also check the descriptions of the voices you create. The same providers write the conversation titles (section 2). They process the content only to produce the result, under their own terms, which may include temporary retention for security and abuse prevention. The list may change as models change; this page will be updated.
- Fish Audio (United States): synthetic voice for most of the translations. For the standard voices, it receives the translated text to be spoken through OpenRouter. For the voices you create (section 2), our server calls it directly: it receives the voice instruction rewritten from your description, keeps the model of the voice you save until it is deleted, and receives the translated text to be spoken in that voice; if the direct call does not answer in time, the same phrase goes to Fish Audio through OpenRouter, with the same voice. We send this data only to create the voice and speak your translations; Fish Audio handles it under its own terms, which also allow it to use it to develop and train its AI models (and what it has already used may not be fully deleted).
- RevenueCat (United States): subscription management.
- Apple and Google: payments and subscriptions, under their own privacy policies.
- PostHog (United States): usage statistics of the app and of this website.
- Cloudflare (United States): hosting of this website.
- Fly.io (United States): server hosting; Upstash (via Fly.io): temporary counters, cache, consent records, the record of the voices you create (without the description or the name) and, for up to 24 hours, the rewritten version of the description; Tigris Data (via Fly.io): storage for "Help improve the AI", for the voice cache and for the previews, samples and reports of the voices you create.
Except for Fish Audio in the voices you create (see above), we require these providers, by contract or under their terms of use, to protect the data in the same way as described in this policy.
We do not sell or rent personal data, do not share it for advertising, do not use advertising identifiers (IDFA/AAID) and do not track you across other apps or websites.
5. How long we keep data
- Audio and text for translation, and the text sent for a conversation title: only while the request is being processed (seconds).
- Conversations: only on your device (up to 200; past that, the oldest one is deleted), until you delete them or uninstall the app, and in the device backup as described in section 2.
- Limit counters: per-minute counters for about 1 minute; daily counters (including the per-network free-phrase limit) for about 36 hours; free-phrase counters (per anonymous ID and per device hash) until about 13 months after the last use.
- Cached subscription status: up to 24 hours.
- Voice cache: the synthesized audio of a repeated phrase is deleted after 75 days without use (at most 180 days in any case); the keyed counters of different people expire after 30 days; the link between a translation and the cached phrases it used, after 30 minutes.
- Voices you create: the original description is not kept; the rewritten English version (and the short summary the app shows), up to 24 hours, and never in the logs; the previews, up to 24 hours; the voice model (at Fish Audio) and the voice's samples and record (on our server), until you delete the voice or ask us to delete your data, or after 90 days without a subscription; the monthly creation counters, up to 40 days; the count of refused requests, 24 hours; reports (reason, anonymous ID and reported synthetic audio), 90 days.
- Technical metadata of each translation, for your ratings (models, languages, timings, estimated cost, plan; no content): 7 days — only if usage statistics are on.
- Usage statistics (app and website): up to 24 months; session replays (app only): up to 90 days.
- "Help improve the AI" data: until you delete it, and at most 3 years; the consent record, for as long as needed to prove your choice.
- Application logs (pseudonymous ID and technical data, no content and no IP): a few days.
- Support messages: for as long as needed to answer you and meet legal obligations.
6. International transfers
The providers above may process data outside your country, mainly in the United States. We rely on the safeguards allowed by the LGPD (such as standard contractual clauses) and the GDPR (such as the EU Standard Contractual Clauses).
7. Legal bases
Performance of the contract (translating, providing the free plan and the subscription and its features, such as the voices you create), legitimate interests (security, fraud and abuse prevention — including the device identifier and IP address — and product statistics, which you can turn off at any time in the app, or with Do Not Track / Global Privacy Control on the website), consent ("Help improve the AI", which you can withdraw at any time) and compliance with legal obligations.
8. Your rights (LGPD / GDPR)
You may ask for confirmation and access, correction, anonymisation, blocking or deletion, portability, information about the entities we share data with, withdrawal of consent and objection, and you may complain to the data protection authority (ANPD in Brazil, or your EU/EEA authority). Many of these you can do yourself in the app: turn off statistics, delete the "Help improve the AI" data, delete your conversations and delete the voices you created. For anything else, write to contato@vrosa.com.br. Because there is no account, include the anonymous ID shown in the app (Settings › About) so we can find your data. We answer within the legal deadlines (15 days under the LGPD; one month under the GDPR).
How to withdraw your permission to send content to the AI companies: in the app, Settings › AI translation consent › Revoke consent. After that, your voice and your text are no longer sent to our server or to the AI providers (section 4): online translation, the translated voice and the conversation titles stop until you accept again (in the same place, or in the sheet the app shows when you try to translate), and offline translation, which runs only on the device, keeps working. There is nothing to delete for past translations, because we do not keep your voice or your text, except as described in section 2; the "Help improve the AI" data and the voices you created stay until you delete them (sections 3 and 2). To withdraw only the permission to create voices, tap "Revoke" on the "Creating voices" line in the same place: the next creation asks for it again, and the voices you already saved keep working until you delete them (Settings › Voice › Voices › "⋯" › Delete voice).
9. Security
Encrypted connections (HTTPS/TLS) only, short-lived access tokens, identifiers stored as keyed hashes, rate limits, and no content in our logs.
10. Children
The app is not directed to children under 13 and we do not knowingly collect their data. If you believe a child has used the app, write to us and we will delete the related data. We do not create children's voices: such requests are refused.
11. Changes
We will update this page and the date above when this policy changes, and tell you in the app about material changes.
12. Contact
68.216.024 Victor Rosa Lima (MEI), Brazilian company registration (CNPJ) 68.216.024/0001-16, address Rua João Antônio Morato do Amaral, 225, Santa Rita, Piracicaba/SP, CEP 13423-258, Brazil — contato@vrosa.com.br