Wyla

Privacy Policy — Wyla

Last updated: November 16, 2026

Wyla ("the App") is operated by Martin Condet, registered as auto-entrepreneur in France.

Wyla listens to your voice to score your English accent. This page says exactly what happens to each recording, who processes it, and how to delete everything. There is no account, no advertising and no tracking. We do measure how the App is used, in a limited way, with our own service (section 8).


1. Data Controller

Martin Condet Contact: [email protected]


2. What the App records

The microphone is on only from the moment you tap the mic button until the take ends: it stops by itself when you finish speaking, or when you tap again. Nothing is recorded in the background. Recordings are made with the iPhone's own microphone, even when Bluetooth earphones are connected.

Each recording is saved on your device: the day-1 reading, your exam readings (so you can replay "day 1 vs today"), and your lesson attempts. They are stored in the App's private container and never leave your device except for the processing steps below.


3. Accent scoring (Microsoft Azure Speech)

To score a reading, the App sends the audio to Microsoft Azure Speech (pronunciation assessment and speech-to-text) and receives a score per sound and a transcript of what a native would hear. The connection goes through a short-lived access token issued by our server; the audio itself goes directly to Microsoft.

We have not enabled audio logging on this service: Microsoft processes the audio to return the result and does not keep it for us. Microsoft's handling of the data is described in the Microsoft Privacy Statement.

A second check on word exercises (our scoring server)

In the word exercises ("Say it 3 times"), the App also sends the same recording, through our server (section 5), to a scoring server that we operate ourselves. It runs an open speech model (wav2vec2) and acoustic measurements on the recording to check the sound you are practising, and returns a verdict for each repetition.

The recording is processed in memory and discarded as soon as the answer is sent. It is not written to disk, not logged, not used to train any model and not shared with anyone. The scoring server receives the recording with the word and the sound being practised, and nothing that identifies you: neither your IP address nor the random identifier described in section 5. This server is hosted by Hetzner in Finland, inside the EU. If it does not answer, the exercise is scored by Azure alone.


4. Your native twin (ElevenLabs)

During onboarding, your 20-second reading is sent through our server to ElevenLabs to build a temporary copy of your voice and generate one sentence with it. The temporary voice is deleted in the same request, as soon as the sentence has been generated, and the App records whether the deletion succeeded. We never generate anything else with your voice, and we never reuse it.

The generated sentence (an audio file) is stored on your device. ElevenLabs processes the audio under its own Privacy Policy.

You must only record your own voice. Do not submit a recording of someone else.


5. Our server

A small server (hosted on Cloudflare) issues the Azure access tokens, relays the twin request to ElevenLabs and relays word-exercise recordings to our scoring server (section 3). It sees:

  • a random identifier created when you install the App, used to limit the number of requests per device and to label the listening answers described below;
  • your IP address, as with any internet request;
  • timings and error codes, in logs that never contain audio or what you said.

Neither this server nor the scoring server stores recordings or transcripts. The server stores two things, both without any audio:

Listening exercise answers

In the listening exercise ("Which one did you hear?"), you hear a word and pick one of two. When you finish a round, the App sends your answers to our server: for each word, the two words offered, the one that was played, the voice that said it, the one you picked and how long you took, with the first language the App guessed for you and the App version. They are stored under the random identifier above, so that the rounds of one install can be read in order.

We use them for one purpose: to check that the exercise works, that is, whether people hear the difference better after a few weeks. There is no audio, no name and no account, and we do not share them with anyone. They are kept for 400 days, then deleted automatically. Our legal basis is our legitimate interest in checking that the method works.

"I said it perfectly" reports

When you tap that button after a word was scored as missed, the App sends the word, the sounds expected and the sounds heard, and the App version. No audio and no identifier are stored with it. Reports are kept for 180 days, then deleted automatically.


6. Stored on your device only

  • Your scores, ranks, lesson history and streak
  • Your goals and the accent guess (which uses your device language and region)
  • Reminder preferences
  • The recordings and the twin audio described above

The recordings and the twin audio never leave your device except for the processing steps above, and none of this is synced to a cloud account. Two kinds of results do leave your device, without any audio: the listening answers and the reports described in section 5, and the summary of your progress sent to Superwall described in section 7.


7. Subscriptions (Apple and Superwall)

Payments are processed by Apple through the App Store. We do not see your payment details or your Apple ID.

The paywall and subscription state are managed by the Superwall SDK, which receives standard device information (device model, OS and App version, a vendor identifier, locale) and purchase events. See the Superwall Privacy Policy.

The App also sends Superwall a summary of your progress, attached to Superwall's anonymous user id (there is no account, so no name or email):

  • the goals you picked, the first language the App guessed for you, your three weakest sounds, your day-1 score and your current rank;
  • how many lessons you have done and your streak;
  • for the listening exercise: that you finished a round, the pair of sounds, your score out of the number of words, how many rounds you have done, which pairs you have mastered and which voice you find hardest.

We use it for two things: to show a paywall that speaks about your own goals and sounds, and to see in Superwall's dashboard how progress and subscriptions relate. Superwall never receives your recordings, the transcripts, or the word-by-word detail of your answers. Our legal basis is our legitimate interest in offering and improving the subscription.

Superwall also sends subscription events (trial start, renewal, price, country) to our own measurement service described in section 8.


8. Usage measurement (Attribura)

The App sends a small amount of usage data to Attribura, our own measurement service (api.attribura.com):

  • An install identifier. A random code the SDK generates the first time the App runs. It is not your device's vendor identifier or advertising identifier. It is deleted with the App, so reinstalling creates a new one. We also store Superwall's anonymous app user id next to it, so that a subscription can be tied to the same install.
  • The onboarding screens you reached. The name of each step (for example "hello", "read", "twin", "paywall") with a timestamp.
  • The days you open the App. The first time you open it each day, with the date the App was installed, to measure whether people come back after 1, 7 and 28 days.
  • Two key actions. That you completed a lesson, and that you completed an exam, at most once per day each. No content, no recordings and no scores are sent.
  • Subscription events. When you start a trial, renew or buy, Superwall's server sends the event (price, country) to Attribura, joined by the same anonymous user id. The App itself does not report purchases to Attribura.

We use this data only for analytics: to see where people drop out of onboarding and whether they come back. We do not use it for tracking or advertising, we do not sell it, and we do not combine it with data from other companies' apps or websites. The App does not show the App Tracking Transparency prompt because it does not track you. We do not ask how you heard about the App, and we do not collect any self-reported source.

Our legal basis is our legitimate interest in understanding how the App is used and improving it. Attribura runs on our own server, hosted by Hetzner in Germany, inside the EU. We keep the data only as long as it is useful for this analysis, and delete it on request.


9. What we do not do

  • No account or email collection
  • No advertising SDK, no advertising identifier (IDFA), no cross-app tracking
  • No analytics on your recordings. The only results that leave your device are the listening answers of section 5 and the progress summary of section 7
  • No tracking: the usage data in section 8 is used only for our own analytics
  • No sale or sharing of data with third parties beyond the processors named above

10. Notifications

Reminders are scheduled locally on your device. We do not run a push server and do not hold a device token.


11. Children

Wyla is not directed at children under 13. We do not knowingly process a child's voice.


12. Deleting your data

  • In the App: Me › Delete my recordings removes every recording, the twin audio and your progress.
  • Delete the App to remove everything it stored.
  • Our servers keep no recordings, Azure keeps no audio for us, and the ElevenLabs voice is deleted at generation time.
  • The usage data in section 8 is linked only to a random install identifier. Deleting the App deletes that identifier, so nothing collected afterwards can be linked to the old one. To have the earlier data deleted, write to us (section 13).
  • The listening answers in section 5 are linked to a random identifier as well, and are deleted automatically after 400 days. To have them deleted sooner, write to us.
  • The progress summary in section 7 is linked to Superwall's anonymous user id, which is deleted with the App. To have the earlier data deleted, write to us.

13. Your rights (GDPR / CCPA)

You have the right to access, correct, export or erase your personal data, and to object to its processing. Because most data lives on your device, you can exercise these rights directly in the App. For the data in sections 5, 7 and 8, or anything else, write to [email protected]. You may also lodge a complaint with the CNIL (France).


14. Changes to this policy

We may update this policy. Changes appear on this page with a new "Last updated" date.


15. Contact

Martin Condet [email protected]