2.0.6

An iTerm2 dictation fix, multilingual voices that follow the language you write in, and half the download size.

• Dictation works in iTerm2 again. Transcribed text wasn't being inserted there at all — and because Auto-Submit's Enter was delivered, you got an empty command submitted instead. The accessibility write reported success while silently doing nothing, so the app never fell back to the typing path that works. Writes are now verified rather than trusted, on every app. Thanks to @natronite for the report and the diagnosis (#49).
• Multilingual voices follow the language you write in. A supertonic:de:M1 voice used to be a hard pin to German — ask for a reply in English and the model would refuse, citing the instruction. The engine now detects the language of each sentence on-device and speaks it in that language, keeping the same speaker. Kokoro voices are unchanged: a Kokoro voice is its language. (#61, by @hakanensari — closes #40.)
• Pin the reply language if you want one. tts_language, or OW_TTS_LANGUAGE per project, names the language every spoken reply is written in, for any voice, whatever the conversation is in.
• Half the download size — 12.7 MB → 7.1 MB. A text-normalization engine that OpenWhisperer never actually used is no longer linked into the build.
• Audio engine upgraded to FluidAudio 0.15.8: better handling of possessives in spoken replies, plus a batch of speech-recognition fixes.

If you use iTerm2: the fix is deliberately conservative — it only overrides a "successful" accessibility write when it can prove nothing changed, because guessing wrong would type your sentence twice. It could not be reproduced on the machine it was written on, so if dictation still doesn't insert, the log says which path it took and that's worth reporting on #49.

If you use scripts/speak.sh: the blocking /v1/audio/speech endpoint is about 10 dB quieter in this release, because it now returns the model's native level instead of normalizing each clip to full scale. Spoken replies are unaffected — different path, same loudness as before.

This build was replaced. 2.0.6 was first published a few hours earlier from commit 5d9aee2; this release now ships bf2b5e7, which adds the size reduction above and several fixes found in review. If you downloaded the earlier build, re-download — both call themselves 2.0.6.

Install: download the DMG, drag to Applications. This build is ad-hoc signed, so macOS drops the Accessibility and Microphone grants on upgrade — remove and re-add OpenWhisperer in System Settings → Privacy & Security if dictation doesn't respond.