2.0.7 — Translate into English as you dictate

Dictation can now translate into English as you speak.

• Translate as you dictate. Turn on Translate to English and whatever language you speak is typed out in English, in whichever app you're in. It reaches well past coding — sit in a meeting in a language you don't speak and read along. Toggle it straight from the menubar menu, or in Settings → Advanced.
• It uses a second speech model, and only if you ask for it. The fast model the app ships with genuinely cannot translate: ask it to and it hands back exactly what you said, with no error, because OpenAI left translation data out of that model's training. So switching translation on downloads a translation-capable model once (~948 MB) after asking you first. Never turn it on and you never download it.
• Mixing languages works. While translating, the spoken language is detected for you, so switching between English and anything else mid-conversation is fine — English stays English, everything else arrives in English. Your Dictate → Language setting is left untouched and applies again the moment you switch translation off.
• Dictation is slower while translating — roughly twice as long per phrase, since the translating model is the larger one (32 decoder layers against turbo's 4). With translation off, dictation is exactly as fast as it was.
• "Delete models" now frees everything. The compiled Neural Engine cache the installed app builds — a few hundred MB — sat at a path the cleanup never looked at, so it was neither counted nor reclaimed.

The DMG was replaced shortly after release. The first build could wedge if you turned translation on and straight back off again while it was still loading — dictation would then stall behind a model load nobody asked for and time out. Fixed in the build now attached. If you downloaded within the first hour, re-download.

Install: download the DMG below, drag to Applications. If macOS blocks it:
xattr -d com.apple.quarantine ~/Downloads/OpenWhisperer-2.0.7.dmg