Last updated: 21 September 2026
This policy explains what AnyAccent records, why, how long we keep it, and what you can ask us to do about it. It is written to be read, not to be survived.
The short version. To tell you how you pronounced something, we have to listen to it, so every practice recording is sent to our server and scored there. From version 1.7 we also keep those recordings and use them to make the coach better. We never use them to work out who you are, we never sell them, and you can have every one of them deleted.
A note on versions. Some of what follows arrives in version 1.7, and is marked. If you are on an earlier version, only the parts not marked that way apply to you: your recordings are sent to our server to be scored and then discarded, unless you switched on Help improve AnyAccent in Settings, in which case they are kept.
AnyAccent is made by Yadu Bhati. For anything in this policy, including a request to see or delete your data, write to yadubhati19@gmail.com. We answer within 30 days, usually much sooner.
When you tap the microphone and speak, the app records you, sends the recording to our server, and scores it there. The scoring model has to hear the audio — there is no version of this product that works without it.
From version 1.7 we keep those recordings. They are the only way we can tell whether the coach is right, find the sounds it judges badly, and improve it. Recording is not optional, because the recording is the product; what is optional is whether yours is used for improvement, and whether we keep it at all. See section 6.
Before version 1.7, recordings were used to produce your score and then discarded, unless you turned on Help improve AnyAccent in Settings, which was off unless you chose it.
Stored with each recording: the sentence you were given, the scores it received, roughly how long it was, the date, and your app’s random installation ID.
When you start the app we ask which language you grew up speaking. It tells you which sounds usually take a speaker of that language the most work, and it labels recordings so we can see, for example, that the coach is weaker on Tamil speakers than on Polish ones.
We also store your country, taken from your device’s Region setting in iOS. This is not location data: it is the country you chose in Settings, it is never finer than a country, and we do not use GPS, Wi-Fi positioning or your IP address to work it out. (Google Analytics does estimate a rough location from IP addresses for its own totals; see App usage and crashes below.)
You can answer “Prefer not to say” to the language question. We would rather have a blank than a guess.
The app makes up a random ID the first time you open it and sends it with each request, so your scores can be compared against your own history rather than against strangers. It is not your name, your email, your Apple ID, your phone number, or your device’s advertising identifier, and it is not linked to any account, because AnyAccent has no accounts. Delete the app and the ID is gone for good.
If you use the report button on a practice screen, we store the reason you picked, the sentence, the scores, the full per-sound result, and any note you type. Please do not put anything private in that note — it is free text and we read it.
Like any web server, ours records each request it receives: the time, what was asked for, and the IP address it came from. We use this to keep the service running and to stop abuse, not to identify you or to work out where you are, and it is not joined to your recordings. These logs are kept for up to 30 days and then deleted.
We use Google Firebase Analytics to count things like how many lessons are finished and where people give up, and Firebase Crashlytics to receive crash reports. These see an app-instance identifier generated by Google, the device model, the OS version, and the events themselves. Google also works out a rough location, at country or city level, from the IP address the app connects from; we only ever see it as totals. They do not see your recordings. We do not use advertising identifiers (IDFA), we do not track you across other apps or websites, and there are no advertising networks in this app. Earlier versions had no analytics of any kind.
If the app cannot reach our server, it can still give you a rough estimate of a take using the speech recognition built into your iPhone by Apple. That audio is processed by Apple under Apple’s own privacy terms, not sent to us, and the estimate is clearly marked as offline. It is not the AnyAccent coach and can be well off.
If you installed AnyAccent after tapping an ad in the App Store, Apple tells us which ad campaign it was (Apple Search Ads attribution). This is a campaign number, not anything about you: no advertising identifier, no tracking across apps, and no permission prompt is involved. We use it only to see which ads are worth paying for.
If you allow notifications, Firebase Cloud Messaging holds a push token for your device so we can send you occasional announcements. Turn notifications off in iOS Settings, or turn “News and tips” off in the app, and it stops.
Subscriptions are sold and charged by Apple. We never see your card, your billing address, or your Apple ID. We receive only an anonymous receipt confirming that a subscription is active.
Your name, streak, progress, unlocked lessons and settings live on your device. If you have iCloud backup on, Apple backs them up under Apple’s terms, not ours.
Your recordings are scored by software, automatically. But a person may listen to them — that is us, checking why the coach got something wrong, or picking examples to train a better model. We keep this to what the work needs, and it is one more reason the delete button in section 6 exists.
Every score in AnyAccent is produced by a machine-learning model. Nobody checks its verdicts before you see them.
It makes mistakes. It mishears sounds, it marks good pronunciation as poor and poor pronunciation as good, and it is less reliable on a noisy recording, a distant microphone, a very short clip, or an accent unlike the ones it was trained on. The scores are a practice aid, not a measurement: they are not a language test, not a qualification, and not evidence of how well anyone speaks English. Please do not treat them as one, and do not let anyone else treat them as one either.
Because of that, the scores have no legal or similarly significant effect on you, so Article 22 GDPR (automated decisions) does not apply. If a result looks wrong, tell us with the report button on the practice screen — those reports are the main way the model gets better.
| What | How long |
|---|---|
| Recordings and their scores | 18 months, then deleted automatically by a job that runs every day |
| Reports you send us | 18 months |
| Server logs (including IP) | up to 30 days |
| Analytics and crash data | Google’s own retention, currently up to 14 months |
| Anything you delete yourself | goes immediately |
From version 1.7, inside the app under Settings › Your recordings:
On any version, writing to yadubhati19@gmail.com gets you the same things: a copy of what we hold, correction, deletion, restriction, or an objection to our using your recordings for improvement. If you are in the EU or UK you can also complain to your national data protection authority.
Your first language cannot be changed in the app, because it labels recordings you have already sent and changing it would mislabel them rather than correct them. Email us and we will fix it properly.
We do not treat your recordings as biometric data under Article 9, because we do not process them for the purpose of uniquely identifying you — see section 4.
Recordings, scores, reports and server logs are stored on a server in Germany (Hetzner Online GmbH). Analytics, crash reports and push tokens are processed by Google (Firebase), which may process them in the United States under the European Commission’s Standard Contractual Clauses and the EU–US Data Privacy Framework. Payments are handled by Apple. We use no other processors.
AnyAccent is for people aged 16 and over. We do not knowingly collect anything from children under 16. If you believe a child has used the app, write to us and we will delete everything we hold.
Traffic between the app and our server is encrypted with HTTPS. The server is reachable only by key, the admin panel is password-protected and not public, and access to recordings is limited to the people who need it for the work described above. No system is perfect; if something goes wrong that affects you, we will say so.
If we change what we collect or why, we will update this page, change the date at the top, and say so in the app before the change applies to new recordings.
Email: yadubhati19@gmail.com