Model packs v2 (EN/FR top tier)

Top-tier EN<->FR on-device packs: Whisper small (STT), Gemma 2 2B Instruct Q4_K_M (MT), bilingual Piper voices en+fr (TTS). sha256-pinned in ModelManifest.kt. See MODELS.md.

whatyousay v0.1.0 (early access APK)

whatyousay v0.1.0 — early access

On-device, offline speech-to-speech translation for Android. Speech in, speech out, fully on your phone. Nothing you say leaves the device; the only network use is a one-time model-pack download over wifi.

Install

1. Download whatyousay-0.1.0.apk below onto an Android 8.0+ phone (arm64).
2. Open it. Android will ask you to allow install from unknown sources — this is expected for a directly distributed APK (not yet on the Play Store).
3. Launch the app and grab the model pack on first use (over wifi). After that it runs offline.

What works

• STT (Whisper) to MT (llama.cpp GGUF) to TTS (Piper), on-device, single-turn conversation.
• Airplane-mode translation once models are installed.

Notes

• Signed with a self-signed release key for direct sideload. A future Play Store build will use Play App Signing.
• Bundles native libraries for arm64-v8a (phones) and x86_64 (emulator).
• Continuous hands-free VAD loop is the next polish; single-turn is solid.

SHA-256 (whatyousay-0.1.0.apk): d4768fa14ef0a218b3cb7c32079037ffc1510977af434ebde675876de4468e44

WhatYouSay - Demo build (UX preview)

Sideloadable debug build for testing the app UX on a phone. Translation uses DEMO engines (fake STT/MT/TTS) - this previews the real conversation + onboarding/model-download UX, not translation quality. Native on-device translation is a later build. Enable install from unknown sources to install.

Model packs v1

On-device model packs for WhatYouSay Phase 1 (sha256-pinned in ModelManifest.kt). See MODELS.md.

• qwen2.5-0.5b-instruct-q4_k_m.gguf (MT)
• stt-whisper-tiny.zip (STT)
• tts-piper-amy.zip (TTS)
• silero_vad.onnx (VAD)