Problem
Realtime translation calls are hard to demo when translated audio loops into STT, cloud runtime cost is high, and device testing needs repeatable evidence.
Tool Case Study
ArcApps Voice Lab
A bilingual WebRTC calling prototype that separates original and translated audio paths for lower-friction cross-language demos.
Realtime voice prototypeRealtime translation prototypeConditional local demo
Solution
Built a local-PC-hosted FastAPI and WebRTC stack with split audio routing, translate-only behavior, and acceptance automation for bidirectional playback.
Result
Verified the workflow with browser E2E acceptance and packaged an Android WebView APK for controlled demonstrations.
How it works
01
Voice input is captured from a browser or WebView client.
02
The audio pipeline separates original audio from translated output to avoid feedback loops.
03
The visual flow is voice input to STT to translation to TTS output, without running the real realtime stack on the homepage.
Build proof
Browser E2E acceptance and diagnostics were prepared for bidirectional translation checks.
Local worker and WebRTC routing were used for controlled demonstration constraints.
Heavy APK, MP4, and realtime runtime artifacts stay external to the website bundle.
Current status
Conditional local demo. This page presents a lightweight case study and links outward for repositories, releases, demos, or heavy artifacts instead of bundling them into the website runtime.
Terminology
voice inputSTTtranslationTTS outputrealtime connectionWebSocket previewlatencylocal workerresident workercold startwarm render
Stack
FastAPIWebRTCOpenAI RealtimePlaywrightAndroid WebView