Real-time translation

Real-time translation converts speech from one language to another while a person is still talking, streaming the result with only a fraction of a second of delay. Unlike batch translation, where you submit text and wait, real-time systems process audio continuously so both sides can hold a normal back-and-forth conversation.

It chains three steps — speech recognition, machine translation, and speech synthesis — and optimises each for latency rather than waiting for a full sentence. SimulSpeak targets sub-second added latency so translated calls feel like ordinary calls.