Live Demo: livecalltranslator.netlify.app
Cross-language voice chats lose momentum when translation lags or lives outside Discord. This project aims to bring real-time translation natively into Discord voice channels with a focus on low latency and seamless integration.
By leveraging Deepgram's streaming STT and Discord's inline bot capabilities, translations appear directly in the chat alongside the original speech—no external tabs or apps required.
The Problem
- Cross-language voice chats lose momentum when translation lags or lives outside Discord
- Existing solutions require context switching to external apps or browser tabs
- High latency makes real-time conversation feel unnatural and disjointed
Goal: Native feel, no context switching, clear in-channel visibility with sub-second translation response.
Architecture
The system uses a streaming STT service (Deepgram) that feeds a translation worker. An inline bot posts both interim and final translations directly to the chat, keeping everything native to Discord.
Technical Constraints
- <250ms interim response — Users see near-instant feedback that speech is being processed
- Sub-second final translations — Full translations arrive quickly enough for natural conversation flow
- Native Discord integration — Works inside Discord chat; keeps moderator controls intact
Reliability & Resilience
- Reconnection guards — Automatic reconnection with exponential backoff on connection loss
- Jitter buffers — Smooth temporary network dropouts without interrupting user experience
- Rate-limit backoff — Respects Discord's API limits while maintaining responsiveness
- Interim/final dedupe — Reduces chat spam by intelligently merging rapid successive updates
Privacy & User Control
- Per-user language preferences — Each user sets their target language individually
- Opt-in by default — Only opted-in voices get processed; users control their participation
- Channel-level toggles — Overlay can be toggled per channel for high-visibility events
Status & Next Steps
Currently in active development. Working on:
- Hardening reconnect paths for edge cases
- Tuning translation consistency across language pairs
- Building a room-health console for monitoring