Privacy

BecomeBulgarian is a language-learning app operated by Lion MGT LLC, 5195 Hampsted VCW #232, New Albany, Ohio 43054, USA. We are the data controller for the information described below. You can reach us at [email protected].

Our representative in the EU. We are based in the United States, so we have named a representative in the EU under Article 27 of the GDPR: Philip Cutting, Mladost 1A, Block 535, Entrance A, Office 1, 1729 Sofia, Bulgaria, email [email protected], phone +359 893 011 945. If you live in the EU or EEA, you may contact our representative about any data-protection matter.

BecomeBulgarian is for adults. You need to be 18 or over to create an account, and everyone who learns on a profile of your account must be 18 or over too. When you create an account you confirm that you are 18 or over and that you agree to our terms and this policy, and we keep a record of that.

• Account details — your email address, your display name if you give one, and your sign-in identifiers (a password we store only as a hash, or the identifier Apple or Google gives us when you sign in with them).
• Profiles — the account holder can add profiles for other adults in the same household, up to three profiles in all on the free plan, the holder's own included. For each profile we keep the name or nickname it is given and the picture chosen for it from the app's own set (never an uploaded photo). Each profile's sentences, recordings, progress and settings are kept separately and belong to that profile. Profiles are not locked from one another: anyone who can open the account can switch to any profile on it, and the account holder's “Download my data” copy includes every profile.
• Your choices and when you made them — that you confirmed you are 18 or over and accepted the terms and this policy, whether each profile allowed its sentences to go to the AI services named below, whether it offered its sentences for the shared course, and the wording version and time of each choice.
• Sentences you write or dictate — the personal sentences you save so the app can translate them, generate audio, and build practice sessions from them, together with what is made from them: the Bulgarian, the transliteration, the English, the explanation, the pronunciation notes and the audio clip.
• Recordings of your own voice — only if you use “Record my own voice”, where you record yourself saying one of your sentences to compare it with the native audio. We keep up to three takes per sentence in private storage. You can play them, and so can anyone who can open your profile on this account; the account holder's “Download my data” copy includes them. A teacher hears one only if you attach it to a teacher review. They are never sent to an AI provider or used to train anything. Delete a take, the sentence or your account and the recording goes.
• Teacher reviews you ask for — the sentence you send, your note, and the teacher's answer. The teacher is a person: a Bulgarian teacher who works for us under a confidentiality agreement, and the app says so before you send. The teacher reads the sentence and your note and listens to any recording you attach; they are shown a pseudonym, never your name or email.
• Learning activity — what you practised, when, and how it went, including the stories and tongue twisters you listened to or tried, so the app can sequence what comes next.
• Story concerns — if you report a concern about a story (it gets the culture wrong, it is offensive, a language mistake, or something else), we keep the reason you chose, any note you add, which story it was, and which profile on your account sent it, so our team can review the story. Only our administrators see it; it is deleted when you delete that profile or your account.
• Server logs — our own server records each request the app makes: the time, the address it asked for, the result and your account id, so we can keep the service running and spot abuse. The app itself contains no crash-reporting or analytics SDK: it sends us nothing beyond the requests you cause by using it.
• Website analytics — on the public pages of becomebulgarian.com we may count page views with Umami, which runs on our own server: cookieless, aggregate counts only. No cookies, no advertising identifiers, no cross-site tracking, and nothing that identifies you. It is never used inside the app or on your account pages.
• Beta list — if you join the waitlist on our website: your email, the platform you chose and the time you gave consent, kept until we have sent the beta link or you ask us to delete it. If you arrive through a contributor's link or enter their code, we record that code with your sign-up so we can thank them. The contributor only ever sees totals, never who signed up. The code travels in the link and the form; nothing about it is stored on your device.
• Contributor applications and assessments — if you apply to work with us, or do paid test work or ongoing work for us: see Contributor applications and AI-assisted assessment below.

When you dictate a sentence or practise saying one aloud, the app hands the audio to your phone's own speech recogniser — Apple's on iOS; on Android, the one your phone uses, usually Google's. Where the phone has Bulgarian installed for offline recognition it stays on the device; where it does not, the phone sends the audio to its maker's service to transcribe, under their privacy policy rather than ours. Either way that audio never reaches us. From dictation we receive only the text that comes back, and only once you save the sentence; from speaking practice, only a score. (Recordings you make with “Record my own voice” are different — you choose to keep them, and they are stored as described above.)

Please avoid putting information you would not want processed into your saved sentences, and do not put other people's personal details in them. They are ordinary personal data to us, but you decide what goes in them.

• To provide the app — your account and profiles, translating your sentences, generating audio, keeping the recordings you make of your own voice, and tracking your progress. Legal basis: performance of our contract with you.
• Sending your sentences to AI services — before a profile's first sentence, the app asks for its explicit permission, names the services and says what each one receives. Nothing is sent until you tap Allow. (App versions from before this screen existed did not ask; they stop adding sentences once the update is required.) If you choose Not now, or later turn it off in Settings, adding sentences stays off for that profile and nothing more is sent; the rest of the app keeps working. Legal basis: performance of our contract with you (it is the feature you ask for), with your permission asked first.
• Optional extras — having a teacher review a sentence (you choose each time), and letting sentences you wrote be used as examples in the shared course (a switch in Settings, off unless you turn it on; each profile decides for itself). Legal basis: your consent. You can withdraw it at any time, and the app works without it.
• Website analytics — cookieless, aggregate page-view counts on the public website. Legal basis: our legitimate interests in knowing which pages are read.
• Thanking contributors who refer you — the contributor code on a beta-list sign-up. Legal basis: our legitimate interests in thanking the people who tell others about us, and in checking that those thanks are not abused.
• Security and abuse prevention — legal basis: our legitimate interests in keeping the service working.
• Keeping a record of your choices — the 18+ confirmation, your acceptance of the terms, and your permissions. Legal basis: our legal obligation to be able to show consent, and our legitimate interest in showing who agreed to what.
• Keeping a contributor application for future work, or staying in touch — only if you ask. Legal basis: your consent, which you can withdraw at any time. See Contributor applications and AI-assisted assessment below.

We use these providers to run the app. They process data only on our instructions, under a data processing agreement or the provider's data processing terms. The AI services receive the words of a sentence, never your name, email or account.

• Our own server in Bulgaria — the app, the website, the database, and the meter that counts what you use (Lago, which we run ourselves).
• Cloudflare — carries traffic between your device and our server and protects it from attacks. It sees requests and their content in transit, including your IP address.
• Amazon Web Services, in Frankfurt, Germany — stores the audio of your sentences, the recordings of your own voice and our encrypted database backups, and sends our emails (the password-reset link).
• Google Cloud — Vertex AI (Gemini), in the EU (the Netherlands) — turns your sentence into Bulgarian, translates it back to check the meaning, spots duplicate sentences, and writes pronunciation notes. For these language steps our system sends your sentences only to AI models located in the EU.
• Anthropic's Claude, through Google Vertex AI or Amazon Bedrock, in the EU — may be used as a second model that cross-checks a translation, or when Gemini is unavailable.
• DeepL (DeepL SE, Germany) — checks translations, but only when we use DeepL's paid API, whose terms allow personal data. While we are on DeepL's free plan, your sentences are not sent to DeepL.
• ElevenLabs (Eleven Labs Inc., United States) — makes the audio of the Bulgarian with its voices; this is the voice service we use today. It receives only the Bulgarian text, in the United States, under its data processing addendum, which is part of its terms. Our account has opted out of letting ElevenLabs use what we send to improve its models, so it does not train on your sentences.
• Google Cloud Text-to-Speech — makes the audio instead of ElevenLabs when we choose one of Google's voices. It receives only the Bulgarian text. Today we use Google's global endpoint, so this step may be processed outside the EU.
• Our teachers — people who work for us under a confidentiality agreement, only for a sentence you send for a teacher review, as described above.
• Apple and Google — if you sign in with Apple or Google, they tell us your email address and an identifier. Your phone's speech recogniser is theirs too, as described above; for those, their own privacy policies apply.

Transfers outside the EU. Making the audio — ElevenLabs, and Google's speech-synthesis step — involves processing outside the European Economic Area, including in the United States. For Google, the transfer is covered by the European Commission's Standard Contractual Clauses in Google Cloud's data processing terms; for ElevenLabs, by the Standard Contractual Clauses in its data processing addendum. We choose EU-based processing where it is practical, but we do not guarantee that all processing stays within the EEA.

We do not sell your data or share it for advertising.

We do not train AI models on your sentences or your recordings. Under their terms with us, Google Cloud (Vertex AI and Text-to-Speech), Anthropic's models through Google or Amazon, and DeepL's paid API do not use what we send them to train their models. Google may keep a request in memory for up to 24 hours to answer faster, and may log a request its systems flag as possible abuse for a limited time. ElevenLabs does not train on what we send either: we have turned off its setting that lets it use our content to improve its models. Under its terms it may keep request text for a limited time. We send it nothing but the Bulgarian text, never your name, email or account.

We keep your account, your profiles and your saved sentences for as long as your account is open. Delete a sentence and it goes, with its audio and your recordings of it. The account holder can delete another profile at any time; that deletes its sentences, recordings, teacher reviews, story concerns, practice history, streaks, statistics and preferences.

If a sentence of yours was copied into the shared course (only possible if you turned that on), the copy has no name attached and stays in the course after you delete your sentence.

The account holder can close the account from inside the app — Settings, then Delete account — or on this website under Account, Your data. Closing it removes every profile, your sentences and their audio, the recordings of your own voice, your teacher reviews, your practice history, streaks and statistics, your preferences, your consent records, anything you sent us through support, and the sign-in itself. It happens straight away and it cannot be undone.

Two things outlive that, and we would rather say so than let you find out later. Encrypted nightly database backups roll off on their own schedule — about 120 days at the latest — so a closed account can persist in a backup until the last one expires; we do not restore a backup to bring an account back. And if you have signed an agreement with us or done paid contributor work for us, the agreements you signed and the records of what you delivered and what we paid have their own retention basis and are kept for the periods set out under “Contributor applications and AI-assisted assessment” below. In that case we anonymise your sign-in instead of deleting it: the email address, display name and password are destroyed and the record no longer points at a person.

Deletion is permanent. Once something is deleted — by you, or by the schedules on this page — it is gone from our file storage within about two days, and nobody can bring it back, not even us. So if you want a copy, download it first: on this website under Account, Your data, or in the app under Settings — both have “Download my data”.

If you are in the EU or UK you can ask us to give you a copy of your data, correct it, delete it, export it, or restrict or object to how we use it. Where we rely on consent, you can withdraw it at any time without affecting what we did before. Write to [email protected] and we will respond within one month.

A copy needs no request either. “Download my data” — on the website under Account, Your data, in the app under Settings on the account holder's profile, and in the Studio — gives you a ZIP file straight away: everything tied to the account in one machine-readable file, every profile on it, your sentences' audio and your own recordings of them, and, if you work with us, your application, recordings, signed agreements and handwriting scans. A few things are not in it — what reviewers wrote about an applicant, for example — and the file's README says what they are and how to ask us for them.

Erasure needs no request: the account holder can close the account in the app (Settings, then Delete account) or here on the website, and it happens immediately. Someone learning on another profile can ask the account holder to delete that profile, or write to us. Ask by email instead if you would rather, and we will do it for you.

You can also complain to your local data protection authority if you think we have got something wrong.

This section is for people who apply to work with us — voice artists, teachers, coaches, editors, evaluators, translators — whether on this website or in the Contributor Studio, and for contributors who do paid test work or ongoing work for us. The same information, in short, is shown below the application form and in the Studio.

На български: кратко резюме на този раздел има под формуляра за кандидатстване и в Contributor Studio, а абзацът за автоматизираните инструменти за записите и частта „Колко дълго пазим данните“ са по-долу и на български. Ако искате целия текст на български, пишете ни на [email protected].

What changed in this version. The single 12-month period for people we do not take on is replaced by the schedule under “How long we keep it” below: 6 months if we never paid you, 5 years if we paid you for any work (a paid test included), an optional talent pool and an optional contact list, and a short record that we met, kept for 3 years after our last contact. If you applied before we published it, we keep to the 12 months we promised you then, unless we have written to you about the change and you did not object.

What we process.

• Your application — your name and email, the languages you hold and your region, the English level you state, the roles and kind of engagement you want, your message, experience and availability, and any links you give us.
• Your test work and delivered work — recordings and other work you deliver under a paid test or an engagement with us.
• Assessments of you — blind evaluations against a published rubric (scores, a verdict, written strengths and concerns, and any red flags), hiring reviewers' scores and notes, ratings and notes on your recordings, notes by our staff, and, if we interview you (by phone, video or in person), the interviewer's scores against a published interview rubric, any red flags, and their written notes. Also the qualification score described below, and the notes automated tools make about your recordings, with the reviewers' checks of them (below).

If we invited you to apply. If we found you on Upwork, on LinkedIn or through a profile or CV you sent us, we may start your application for you instead of asking you to type it all again. We copy, by hand, the text of your profile as it is shown to us there — your work history, the languages you speak, your English level and the kind of work you do — with your name, the email address you gave us and the link to the profile. Nothing is collected automatically: we do not scrape Upwork or LinkedIn. We then pre-fill the application from that text ourselves, or with the help of the EU AI model described below: before the text is sent, names, email addresses, links and phone numbers are removed from it, the model is told to use only what the text says and to leave everything else empty, and its draft is not kept anywhere but in the application we then check and save. We do this to save you time, on the basis of our legitimate interest in finding contributors (Art. 6(1)(f) GDPR), and for no other purpose.

We send you a personal, single-use link. The first time you open it you are told, before anything is asked of you, that the application was copied from your profile, and you are shown exactly what we copied. You can correct or delete any answer, and delete the copied profile text itself, when you confirm; nothing becomes your application until you do. It is kept, and deleted, exactly like any other application (“How long we keep it”, below). If you never confirm it, we delete what we copied no later than 6 months after we copied it — sooner if you ask.

Why, and our legal basis. To decide whether to offer you work, to run the paid test, and to decide whether to continue working together. Legal basis: taking steps at your request before entering into a contract with you, and performing that contract once it exists (Art. 6(1)(b) GDPR); and our legitimate interest in choosing and keeping contributors carefully and fairly (Art. 6(1)(f)), which also covers the AI-assisted part described below. We inform you about that part here, below the application form and in the Studio.

• Talent pool — keeping your application, test work and assessment for future work, only if you ask. Legal basis: your consent (Art. 6(1)(a) GDPR; for candidates in Bulgaria also Art. 25k of the Personal Data Protection Act, where it applies). You can withdraw it at any time.
• Staying in touch — keeping only your contact details and the record of when and why we were in touch, so that we can offer you work, only if you ask. Legal basis: your consent, which you can withdraw at any time.
• A record that we met — described under “How long we keep it”. Legal basis: our legitimate interests (Art. 6(1)(f)).
• Records of paid work and of signed agreements — to show what was agreed, delivered and paid for, and to establish, exercise or defend a claim under the contract. Legal basis: our legitimate interests (Art. 6(1)(f)).

How the assessment works.

• People's ratings come first. Blind evaluators, hiring reviewers, whoever interviews you, and the people who rate recordings give scores. We combine them into a qualification score out of 100, giving the structured blind evaluations the most weight, and we record how many people rated you and how far they agree. An interview is not blind — the interviewer knows who you are — so its score is kept apart from how far the other raters agree.
• An AI system reads the written material. It reads the answers in your application and the written notes of the people who assessed you or your work, including the notes of anyone who interviewed you, and returns a short summary; judgements on language proficiency, relevant experience, role fit, communication and reliability, each of which may be marked as impossible to judge from text; and, every time, the assumptions it made and what the material cannot tell us. It is not shown the scores people gave, and it cannot listen to recordings: pronunciation, accent and delivery are judged only by people. Two models do this together: one answers a fixed set of yes/no and rating questions that give the numbers, and the other writes the summary and the reasoning. Where they disagree, the founder is shown both.
• Its effect is bounded and advisory. Its reading can move the score from people's ratings by at most 10 points out of 100. Before anyone has rated you, its reading alone is capped at 40 and marked provisional, so it can never make anyone look strong on its own.
• What the AI is not given. Your name, email, phone number, links, location or region, your rate, the English level you stated, the scores and verdicts people gave, our internal notes and our decision. Names and contact details that appear inside the text are removed automatically before it is sent — which is best effort, so the AI is also told to ignore identity. Nothing you write about yourself in the Studio, other than your application, is given to it.
• Automated tools on your recordings, on our own computers. We may run automated tools on your recordings on our own computers — speech-to-text and audio analysis — to split a recording into its parts and compare it with the script: which parts were read, where the words heard differ from the script, and technical facts such as the file's format and levels. Nothing is sent to another company for this. People check the results, and the tools decide nothing on their own: the results are notes to check, not conclusions, and hiring reviewers may be shown some of them, marked as unverified, to check by listening. They are not given to the AI system described above, which still cannot listen to recordings, and they are kept and deleted with the rest of your assessment.
• Checking individual pieces of work is separate. We may use AI to check the quality of individual pieces of work — a phrase, a translation — as quality control of the content. That checks the piece of work; the qualification score of you stays driven by people's ratings, with the AI reading above as advisory only.
• A person decides. The score is a decision aid that only the founder sees. The founder makes every decision — whether to offer you work, to continue it or to end it — and nothing changes your status, your work or your pay automatically. We make no decisions based solely on automated processing.
• When it runs. The score is updated as new ratings and notes arrive while your application and any paid test are being considered. If you work with us, the same kind of assessment may be applied to the work you deliver.

Where it is processed, and by whom. Applications, test work and assessments are stored on our own server in the EU. The AI step uses two processors. Both receive the same material, after names, email addresses, phone numbers and links have been removed from it as described above.

• Google writes the summary, the judgements, the assumptions and the limits, using Google Cloud's Vertex AI in Google's EU region in the Netherlands; our system only sends this material to Google at an EU endpoint. Google acts as our processor under the Google Cloud Data Processing Addendum and does not use the material to train its models. Google may keep it in memory for up to 24 hours as a cache to answer faster, and if Google's systems flag a request as possible abuse, it may be logged by Google for a limited period to investigate that.
• TypeSafe (TypeSafe AI, Inc.) processes the material in the United States. It answers the fixed yes/no and rating questions — for example, whether the notes describe native-level Bulgarian — and our system turns those answers into the numbers of the AI reading. TypeSafe acts as our processor under its data processing agreement, which incorporates the European Commission's Standard Contractual Clauses for the transfer to the United States, and does not use the material to train its models. It runs on its own US cloud and computing providers. It may keep the material for as long as reasonably necessary to provide the service, to monitor for abuse and to comply with the law, and it may keep technical records derived from it, such as logs and statistics. We use TypeSafe only for contributor applications and assessments, never for the data of people learning with the app.

How long we keep it.

• If we do not take you on and have never paid you: your application, test work (recordings and other files), ratings, notes, evaluations, interview notes and qualification score — including the AI reading — are deleted no later than 6 months after our decision, sooner if you ask. A decision to keep you in mind without offering work now is a decision. An application we never took further is deleted 6 months after it was last active.
• If we paid you for any work — a paid test or work sample included: your work (recordings and other deliverables) and the records about you — your application, the assessment, ratings, notes and interview notes, and the contract and payment records (the Upwork contract, the dates and amounts, and any agreement you signed with its signing record) — are kept for 5 years after the last payment or the end of the last paid contract, whichever is later: the period in which a claim under the contract is still possible. Each new payment starts the 5 years again, and shortly before they end we ask whether you would like us to keep your contact details. Then they are deleted.
• How we use paid test work: if you have not signed our Voice Recording Licence or Content Contribution Licence, work we paid you to do as a test is used only inside the company, for internal insight. People at BecomeBulgarian, and reviewers and evaluators bound by confidentiality, listen to it and analyse it — to compare takes, to calibrate how people rate, as a quality benchmark, and to improve our briefs and our assessment — and our processors in the EU may measure its quality automatically (for example pronunciation or background noise). It is never published, and never used to train or fine-tune an AI model or to build a voice model, without a separate licence signed by you. Test work we did not pay for is used only to assess you.
• Talent pool — only if you agree: if you ask us to keep your application, test work and assessment for future work, we keep them for 2 years from the day you agree, and ask you again before that ends. If you do not say yes, or if you leave, they are deleted under the rules above.
• Staying in touch — only if you agree: if you ask us only to stay in touch, we keep your name, your contact details and the record of when and why we were in touch — not your application, test work or assessment — for 2 years from the day you agree, renewable in the same way.
• A record that we met: your name, the email address and phone number we have for you, the profile link you applied with, where you came from (for example Upwork), and one dated line for each time we were in touch with you — an application, an interview, a paid contract, a payment, a message — with its purpose in a few words (for example “voice audition”), how far you got, whether we interviewed or paid you, the contract reference and our decision. No scores, notes, answers or recordings. It is kept for 3 years after our last contact with you, so each new contact starts the 3 years again; never for less than 3 years after our decision; if we paid you, at least as long as the records of paid work above; and for as long as a talent-pool or staying-in-touch consent of yours runs. Automated emails from our systems do not count as contact. We use it to recognise you if you apply again, to avoid testing you twice, to keep to Upwork's rule that work with people found there stays on Upwork for 24 months, and to answer any complaint about our decision (a complaint of discrimination can be made for 3 years). Only the founder sees it; it is never shown to reviewers or evaluators and never given to an AI system. Legal basis: our legitimate interests (Art. 6(1)(f)). You can object, and we will delete it unless we need it for a complaint or claim already made.
• If you signed our agreements but we never paid you: the agreements you signed, with the English-level confirmation behind them, are kept for 3 years after our decision — as long as their confidentiality terms outlive the work.
• If we work together: your signed agreements, the records of what you delivered and what we paid, the work you licensed to us and your contact email are kept for as long as we use your work, and then for 5 years. The assessment from when we took you on is deleted 3 years after the work ends.
• If you applied before we published this schedule: we promised then to delete your application, test work and assessment within 12 months of our decision. We keep to that promise for what we collected then, unless we wrote to you about this schedule at least 30 days before that date and you did not object.
• If we met on Upwork: Upwork keeps its own records of our messages and contracts under its own policies; we do not control them, and nothing above deletes anything on Upwork.
• Backups roll off within about 120 days of each of these dates.

Your rights as an applicant or contributor.

• Access — a copy of your application and of the assessment of you: the written comments of evaluators and reviewers, your qualification score, and what the AI summarised, inferred and assumed. Everything you gave us — your application, recordings, agreements and handwriting — you can download yourself in the Studio with “Download my data”. The assessment is not in that download: ask us for it, and we send it within one month.
• Objection to the AI-assisted part — you can object at any time. We then stop using the AI reading for you and assess you on people's ratings and notes alone.
• Human review and an explanation — you can ask for any decision about you to be reviewed by a person and explained to you, including the main reasons and what part the assessment played.
• Correction — you can have inaccurate data corrected, and add your own comment to an assessment you disagree with.
• Erasure — you can ask us to delete your application and the assessment of you at any time. The record that we met goes too, unless we need it for a complaint or claim already made. If you have signed an agreement with us or done paid work for us, the contract records described under “How long we keep it” — the signed agreements with the English-level confirmation behind them, and the records of what you delivered and what we paid — are kept on their own legal basis, for the periods given there.
• Withdrawing a consent — you can leave the talent pool, or stop us staying in touch, at any time: in the Studio, through the link in any of our messages, or by writing to us. We then delete what we would already have deleted without your consent.

Write to [email protected] and we will respond within one month. You can also complain to a data protection authority, as described under “Your rights” above.

BecomeBulgarian is for adults. It is not intended for anyone under 18, and we do not knowingly collect children's data; if we learn we have, we delete it. If you think a child is using the app, please tell us at [email protected].

We will update this page when what we do changes, and change the date at the top. If a change materially affects you, we will tell you by email or in the app before it takes effect.