A high-performance Android application built for instant, hands-free spoken conversations across 19 languages. Features automated turn detection, 180° screen flipping, and streaming neural translation.
A family of high-performance, model-augmented Android applications engineered for speed, utility, and modern mobile architecture.
Real-time face-to-face voice-to-voice translator featuring client-side Silero VAD turn detection, 180° dynamic interface, and NDJSON streaming pipeline.
Minimalist book note capture app combining on-device ML Kit OCR passage scanning with local Whisper.cpp speech recognition.
Self-hosted client-server Whisper flow with GPU backend acceleration for system-wide voice dictation across mobile and desktop.
Designed from the ground up to remove phone friction in cross-lingual face-to-face interactions.
On-device neural Voice Activity Detection automatically senses speech pauses to trigger end-of-turn translation without touching buttons.
Dynamic Compose layout rotates speech card 180° for the person sitting across from you, creating an intuitive two-way table companion.
Audio frames stream to high-throughput Whisper cloud engines, returning clean transcripts with context-aware punctuation.
Neural Machine Translation with fallback routing ensures accurate idiomatic translation for global travelers and conversations.
High-fidelity neural text-to-speech outputs natural human inflections so the recipient hears clear, audible voice responses.
Edge Workers stream transcript chunks, translation blocks, and base64 audio frames back to Android concurrently to minimize end-to-end latency.
A visual tour of the Material 3 Compose interface and audio pipeline.
Dual speaker interface with 180° inverted top screen for person B and upright bottom card for person A. Zero tap requirement during active dialogue.
Built using Android Clean Architecture with explicit separation across UI, domain, data, and edge infrastructure layers.