privacy policy

you create an account to practise speaking english. we store your conversations and your pronunciation scores so you can track progress. your microphone audio is sent straight to microsoft azure to be scored and is not stored by us. we don't sell your data, and we don't run ad trackers.

accent·ai is built by ogbuilds, a uk-based studio. for the purposes of uk gdpr, ogbuilds is the data controller for accent·ai. we process your data in the uk and through the subprocessors listed below.

• account info — your email address, your name (if you provide one), and a securely hashed password. we never store your password in plain text.
• your learning profile — your self-reported english level and the goal you pick during onboarding, used to tailor the tutor to you.
• conversations — the scenarios you practise, the text of what you and the tutor say, and the grammar/phrasing feedback generated for each of your turns. these are saved so you can revisit them and see your progress.
• scenarios you create — if you turn an idea into your own practice scenario, we store it on your account.
• pronunciation scores — when you use “say it back”, we store the reference text you were asked to say and the numeric scores returned (accuracy, fluency, completeness, prosody, and overall). we explain what happens to the audio itself below.

pronunciation practice needs your microphone. it's worth being precise about where that audio goes:

• when you press “say it back”, the recording is captured in your browser and sent directly from your browser to microsoft azure's speech service, which transcribes it and scores your pronunciation. the raw audio does not pass through, and is not stored on, accent·ai's own servers.
• we never receive or keep your raw microphone audio. only the recognised transcript and the numeric scores azure returns are sent back to your browser, and only those scores plus the reference text are saved to your progress.
• azure's handling of the audio it receives is governed by its own terms; microsoft states it does not use this speech data to train its models. the short-lived access token your browser uses is minted by us and expires in a few minutes.

when the tutor speaks back to you, we send the textof its reply (never your audio) to elevenlabs to synthesise a natural voice. if voice synthesis isn't available, your browser's built-in speech voice is used instead.

we use your data only to run accent·ai: to sign you in, to hold a conversation in character, to generate grammar and phrasing feedback on what you say, to score your pronunciation, and to show you your history and progress. we don't sell your data or share it for advertising.

• performance of a contract — to provide the service you signed up for (your account, conversations, and feedback).
• consent — for microphone access: pronunciation assessment only happens when you choose to use it and your browser grants mic permission.
• legitimate interest — to keep the service secure and working (for example, rate-limiting to prevent abuse).

we use a small set of trusted providers to run accent·ai:

• vercel — hosting for the app and its api.
• neon — the postgres database where your account, conversations, scenarios, and pronunciation scores are stored.
• vercel ai gateway— routes conversation and feedback requests to a large language model provider. the text of your conversation is sent to generate the tutor's replies and your feedback.
• microsoft azure speech — receives your microphone audio (directly from your browser) to transcribe and score pronunciation.
• elevenlabs— receives the text of the tutor's replies to synthesise spoken audio.

each provider only receives what it needs to do its job. we don't use third-party advertising or analytics trackers in the app.

we use a session cookie to keep you signed in (set by our authentication layer). your browser may also keep small preferences locally, such as the voice/accent you pick. we don't use cookies for advertising or cross-site tracking.

we keep your account, conversations, scenarios, and pronunciation scores for as long as your account is active so your history and progress stay available. if you ask us to delete your account, we'll remove your personal data. as noted above, your raw microphone audio is never stored by us in the first place.

under uk gdpr you can ask us to:

• access a copy of the data we hold about you;
• correct anything that's wrong;
• delete your account and its data;
• export your data;
• withdraw consent (for example, by revoking microphone access).

to make any of these requests, reach us via ogbuilds.

some of our providers (the ai gateway and its model provider, azure, and elevenlabs) process data on servers outside the uk, including in the united states. where that happens, we rely on the providers' standard contractual clauses and equivalent safeguards to protect your data.

accent·ai isn't intended for children under 13. if you believe a child has created an account, contact us via ogbuildsand we'll remove it.

data is encrypted in transit (https). passwords are stored only as salted hashes. the azure speech token your browser uses is short-lived, and our subscription keys stay on the server and are never exposed to your browser.

if this policy changes, we'll update the date above and, for material changes, flag it in the app.

questions about your data? reach out via ogbuilds, the studio behind accent·ai.