ora

Ora provides real‑time speech translation directly on a Mac equipped with Apple Silicon. It captures microphone audio, detects voice activity, transcribes speech, and translates the text, all using on‑device models that run on the Metal GPU. The process is split into four stages—capture, voice‑activity detection, automatic speech recognition, and translation—so partial captions appear while the user is still speaking and the final translation is committed after a brief pause.

The software is intended for users who need simultaneous interpretation without relying on internet connectivity or external services. Because every inference step runs locally, no telemetry, accounts, or subscriptions are required, and the application works in environments such as airplanes, subways, or remote trails where network access may be unavailable.

Ora distinguishes itself by combining a streaming on‑device language model with Apple’s MLX Swift framework, delivering translations at roughly 600 ms per partial result. It supports multiple source and target languages, handles idiomatic and technical language, and remains free to download and use indefinitely.