Voluble - real-time speech translation
Voluble translates speech to speech in under 3 seconds. You talk, someone on the other end hears you in their language, fast enough to hold a real conversation.
The chain is Deepgram for speech-to-text, DeepL for translation, and ElevenLabs for text-to-speech, wired together over WebSocket streams with backpressure so nothing floods or stalls. The frontend is React and Vite; the backend runs on Express and Bun. There's a companion Flutter app that uses isolates to keep audio processing off the UI thread.
Two of the Rust crates I contribute to, elevenlabs_rs and deepl, sit directly under this pipeline. Voluble got me into the YC S26 applicant pool.