Privacy Policy

Last updated: 14 November 2026

Summary. Live Sound Translator has no accounts, no advertising and no tracking, and we run no analytics of our own. We operate no server for it and receive no data from it. On Mac and iPhone your audio never leaves your device. If you choose a cloud AI provider, transcribed or typed text (and, on Android, speech audio if you choose cloud speech recognition) goes from your device directly to that provider, not to us. The Android app uses Google ML Kit for on-device translation, which sends Google basic diagnostics; see Android.

Live Sound Translator ("the app") is published by VIET NAM CODERS COMPANY LIMITED ("we"). This policy explains what the app does with your data.

Audio

On the Mac, the app captures the audio your Mac is playing, or the microphone when you choose it as the audio source. On iPhone, it listens to the microphone. Either way, audio is processed on your device by Apple's on-device speech recognition. It is never stored and never sent anywhere.

Transcribed and translated text

On-device translators (default). Text is translated on your device by Apple's Translation framework or by the Apple Intelligence model, and does not leave it.
Cloud translators (optional). If you choose OpenRouter, Claude or an OpenAI-compatible service, the app sends the transcribed text, plus a few earlier sentences for context, directly from your device to the service you configured. The app asks for your permission before it first sends text to a service. That text can include anything said in the audio, such as names or other personal details. The service handles it under its own terms and privacy policy, for example Anthropic's. It does not pass through us and we do not store it. Audio is never sent.
OpenRouter. OpenRouter passes the text on to the company that runs the model you chose; both handle it under their own policies (OpenRouter's privacy policy). Signing in happens in your browser on openrouter.ai. The app never sees your OpenRouter password; it receives only an API key, which it keeps in your device's keychain. You can revoke that key at any time in your OpenRouter settings.
A model on your own computer. If you point the app at a server running on your Mac or local network, such as Ollama or LM Studio, the text stays on that computer.
The current session's transcript is kept in memory so you can read and export it. It is discarded when you clear it or quit the app. Files you export are saved only where you choose.

Typed text and reading aloud

On iPhone you can also type text to translate. It is handled exactly like transcribed text: translated on your device, or sent to the cloud provider you chose. Reading text aloud uses the voices installed on your device; nothing is sent anywhere.

Data stored on your device

Your settings, in the app's preferences.
API keys you enter, in your device's keychain. They are sent only to the service they belong to.
The destinations you have allowed the app to send text to, so it does not ask again.

To delete the app's data, remove the app, its container at ~/Library/Containers/vn.webpress.live-sound-translator, and its keychain items (search for vn.webpress.live-sound-translator in Keychain Access).

Android

The Android app works like the iPhone app, with these differences:

Audio sources. It listens to the microphone, or, for live subtitles, to the sound other apps play. Android captures that sound only after you allow screen capture each time you start subtitles. The app uses only the sound; it never records or looks at the screen image. Audio is never stored.
Speech recognition. You choose how speech is recognized:
- Offline models (Vosk) run entirely on your phone. The models are downloaded once from alphacephei.com, which sees an ordinary file download (your IP address) and nothing else.
- Android speech recognition uses the speech service on your phone, usually Google's. With "Prefer on-device recognition" on, it stays on the phone for languages you have downloaded; otherwise Google may process the audio on its servers under Google's privacy policy.
- Cloud speech recognition (optional) sends short clips of speech audio directly from your phone to the OpenAI-compatible service you configured (for example OpenAI or Groq), with your API key, only after you allow it. That service handles the audio under its own policy. Nothing passes through us.
On-device translation. The default translator is Google ML Kit. Text is translated on your phone and is not sent to Google. ML Kit downloads its language models from Google and, like other ML Kit features, sends Google diagnostic and usage information (device and app details, an installation identifier, performance data, error codes and the languages used, but never the text being translated) over an encrypted connection, under ML Kit's terms and Google's privacy policy.
Cloud translators work exactly as described above. API keys are stored encrypted with a key held in Android's hardware-backed Keystore, and are not included in backups.
Conversations and subtitle sessions are saved in the app's private storage on your phone so you can read and share them later, until you delete them in History or uninstall the app. They are included in Android's own device backup if you have it turned on.
Purchases are Google Play in-app purchases. Google processes them; the app only learns from Google Play whether you have bought the lifetime unlock. The 30-day trial is counted on your phone.
Reading aloud uses the text-to-speech voices installed on your phone.
To delete everything the app keeps, delete your conversations in History, or uninstall the app (or clear its storage in Android Settings → Apps).

Purchases

The free trial and the lifetime unlock are App Store in-app purchases. Apple processes them; the app only learns from Apple whether you have them. We receive no payment details.

Speech models and language packs

macOS and iOS download speech models, translation language packs and the Apple Intelligence model from Apple when you request them. Those downloads are covered by Apple's privacy policy.

Identifying speakers

If you turn on "Identify speakers", the app works out who is speaking on your device; the audio still never leaves it. The model it needs is downloaded once from Hugging Face (huggingface.co), which sees only an ordinary file download: your IP address, and nothing about you, your audio or your transcripts. Hugging Face handles that request under its own privacy policy.

Acknowledgements

Speaker identification uses FluidAudio (Apache 2.0) and the Core ML conversion by FluidInference (CC BY 4.0) of NVIDIA's Streaming Sortformer diarization model (NVIDIA Open Model License). The icons on this site are from Lucide (ISC License).

This website

This site uses no cookies, no analytics scripts and no third-party resources. It is hosted on Cloudflare Pages; like any web host, Cloudflare processes technical request data such as IP addresses in order to deliver the pages, under its own privacy policy.

Children

The app is not directed at children. We collect no personal data from anyone.

Changes and contact

Changes to this policy are published on this page and with app updates. Questions: [email protected].