Real-time translation is finally real
Live speech translation hit a turning point: latency dropped below 1 second, accuracy crossed 95% on major language pairs, and three separate models (Whisper-v3, Seamless v3, Google Translatotron 4) shipped production versions in Q2 2026.
The practical floor—where phone calls, live events, and classroom use become seamless—is no longer a research milestone. It's a shipped product.