Wordwave Privacy Stand 14 Nov 2026

Privacy Policy

Your videos and projects are stored on your device. Wordwave has no account, no tracking and no advertising. This is what leaves your device: the sound of a clip, if you allow it and the captions are not written on the device; anonymous signals that count how many people use the app and how its captions were written; and three downloads, a small settings file from likwest.com, the speech model and, for songs, the model that separates the voice from the music.

Camera and microphone

Wordwave does not use the camera or the microphone. It records nothing.

Your videos and projects

Projects, with their captions and looks, are stored inside the app on your device. Wordwave reads a video, a sound file or a subtitle file only when you pick it in Photos or Files, or open it with Wordwave. When you export, the video or the text file goes where you choose: to Photos, to a folder or to the share sheet. Wordwave asks once before it saves a video to Photos. Wordwave does not use iCloud. A project stays on the device where you made it.

Captions on your device

On a Mac, and on an iPhone or iPad with at least 6 GB of memory, Wordwave writes the captions on the device once its speech model is there. Nothing of the video is sent for this, and it needs no permission. The model is about 574 MB. Wordwave downloads it once, over Wi-Fi unless you allow mobile data, from Hugging Face, a company in the United States that hosts it. Hugging Face sees the IP address of that download, as any website does, and nothing about you or your videos. If the model is not there yet, or the device is too slow, too warm or short of memory, Wordwave uses our transcription service below instead, with your permission.

For a song, Wordwave first separates the voice from the music on the device, so that it hears the words better. The model for this downloads once from our server (api.lyrics.likwest.com), which sees the IP address of that download. An iPhone or iPad that cannot run this model skips the step.

When you open Wordwave, it reads a small settings file from likwest.com that says whether captions may be written on the device. Our web server sees the IP address of that request, as it does for every page of this site.

Captions from our transcription service

On an iPhone or iPad with less than 6 GB of memory, and wherever the captions cannot be written on the device, Wordwave needs external AI speech services to write them. It asks for your permission before it sends anything, and our legal basis is that permission. Wordwave remembers your answer on this device. To withdraw it, delete the app; Wordwave then asks again before it sends anything. Withdrawing doesn't reach back to audio already sent. Without your permission, you can still open a subtitle file you already have.

With your permission:

  1. Wordwave sends the sound of the clip you chose, never the picture, over an encrypted connection to our transcription service (api.lyrics.likwest.com), hosted at Hetzner in Finland. The sound can include anything said or sung in the clip. It carries no title, no file name, no name and no account.
  2. Our service sends the audio to Groq, an external AI speech service, which turns it into text. If Groq can't answer, our service sends it to DeepInfra instead, and if DeepInfra can't answer, to fal. Both are external AI speech services too.
  3. The text returns to your device.

What each service keeps:

  • Our transcription service. We delete the audio about a minute after the job ends. The recognised text stays in the server's memory until newer jobs replace it or the server restarts. Our server logs record the IP address of each request, with no audio and no text. The access logs are deleted after 14 days. The application logs are overwritten by newer entries, so they have no fixed period.
  • Groq. We have turned on Zero Data Retention for our account, so Groq keeps neither the audio nor the text. Its terms don't allow it to train models on them.
  • DeepInfra, only if Groq can't answer. DeepInfra's policy says it holds the audio in memory while it works, writes it to no disk, deletes the text once it has returned it, and doesn't train models on either.
  • fal, only if Groq and DeepInfra can't answer. We have told fal not to keep the audio and the text in its request history. That setting covers the request history only. fal's terms allow it to use anonymized or aggregated data derived from them to develop its products and AI models.

Groq, DeepInfra and fal are based in the United States, so your audio leaves the EU. We do not pass your IP address to any of them.

We use the audio only to write the captions, and we do not train anything on it. Apart from what is listed above, we keep only numbers about each job, such as its length, the number of words and the language.

Changing a look by typing

On a device with Apple Intelligence, you can type what you want changed about a look. Apple's language model on your device reads what you typed and the settings of the look, and changes them. Nothing of this is sent to us or to anyone else. Without Apple Intelligence, the field says so and sends nothing.

Translated captions

A translated second line is made by Apple's Translation on your device. Nothing of your captions is sent to us. If a language is not on your device yet, the system offers to download it from Apple.

Anonymous usage counters

Wordwave sends four kinds of signals to TelemetryDeck (EU-hosted), and no other usage data:

  • a session signal when you open the app, with the app version, the system version and the device model, so we can count how many devices use Wordwave each month;
  • an export signal when a video export finishes, with technical details such as the format, the length of the video and how long the export took, as numbers and yes/no values;
  • an access signal when you open the app and when its state changes, with one word for the state of the app on this device (for example "purchased"). A sheet about the purchase sends the same kind of signal when it opens and when it closes, with what opened it (for example export) and how it ended (for example "not now"). It carries no purchase details, no Apple ID and no dates;
  • a captions signal after Wordwave has listened to a clip: whether the captions were written on the device or by our service and why, how long the device took and how warm it got, how much memory it used, whether the model download started, finished or failed, and the device model with its memory class (for example 8 GB).

All signals:

  • contain no personal information: no names, no emails, no file names, and nothing of your videos, your captions or your sound;
  • use an anonymous identifier made on your device, which cannot be traced back to you;
  • do not use the advertising identifier (IDFA) and are not used to track you across apps or websites.

Purchase

The one-time purchase goes through Apple. We don't see your name, your Apple Account or your payment details. Wordwave learns from Apple only whether the purchase was made, and keeps that and the number of exports on the device.

Children

Wordwave is not directed at children under 13 and does not knowingly collect personal data from them.

Your rights

We have no account and no name for you, only the IP address in the logs described above. Deleting a project in Wordwave deletes it from the device. Deleting the app removes its projects and settings from the device, your permission for the transcription service included. Videos and files you exported stay where you put them. You also have the right to access, correct and delete your data and to complain to a data protection authority.

Changes

If this policy changes, we'll update the date above and post the new version on this page.

Contact

info@likwest.com