Travel and communications · Case study
Offline voice-to-voice translation mobile app
A mobile app that translates speech to speech, even without a connection. It chains speech recognition, translation and text-to-speech, uses compressed on-device models for offline language packs, and keeps conversation history in a chat-style view stored only on the phone, with no cloud sync.
The challenge
Translation apps usually stop working without a data connection, which is exactly when travelers need them, and they send every conversation to the cloud. The client wanted a translator that works offline and keeps conversations private.
Our approach
We built a cross-platform React Native app with a speech-to-speech pipeline: speech recognition, translation, then text-to-speech. Compressed on-device models with TensorFlow Lite power offline language packs, and history is stored locally only.
What we built
- Voice-to-voice pipeline (speech recognition, translation, text-to-speech)
- Downloadable offline language packs using model compression
- Chat-style conversation history stored on the device
- Cross-platform mobile app built with React Native and Expo
Highlights
- Offline packs for 10+ languages
- No cloud sync, so conversation history never leaves the phone
Technology
- React Native
- Expo
- TensorFlow Lite
- Google Translate API
- SQLite
The hardest part of an offline translator is fitting the models on a phone. We compressed them for TensorFlow Lite and packaged them as downloadable language packs, so the speech-to-speech pipeline runs on the device once a pack is installed.
Conversations appear as a chat thread, stored locally in SQLite and never synced to a server. Travelers get a translator that keeps working without a signal and keeps what they said to themselves.
Client details are kept confidential. On a call we can walk you through this project in more depth, under NDA.