Coco Voice 0.9.3

If you're on 0.9.2, this update installs itself — this is the first release that exercises the auto-update pipeline end to end.

Coco Voice is a rebranded fork of the open-source Handy dictation app (MIT). The original license is retained.

What's new

• Smarter auto-learn corrections (F2). When you edit a transcription in History, the learned correction now keys off the post-processed text you actually saw — not the raw transcript — so learned pairs reliably match what gets applied on future dictations.
• Redesigned spoken corrections (F3). The old version guessed at corrections from keywords ("no", "actually", …) and pasted the fix after your text. It's now an explicit hotkey: assign a key to Speak Correction in Settings → Post Process → Iterative Correction, press it, and speak your correction — it replaces the last dictation instead of appending. No more false-triggers on normal speech, no more double-paste. Ships enabled but with no key assigned, so nothing changes until you bind one.
• Upgrade safety. Older settings stores that used the legacy coco_keys value now parse correctly, so your settings survive the update.

Install (macOS, Apple Silicon)

Easiest — open Terminal and paste:

curl -fsSL https://raw.githubusercontent.com/coco-research/Coco-Voice/main/install.sh | bash

Or download Coco-Voice_0.9.3_aarch64.dmg below and see INSTALL.md for the one-time Gatekeeper step. (If you're already on 0.9.2, you don't need to do anything — the app will offer this update itself.)

Known limitations (honest notes)

• This build was produced on a Command Line Tools-only toolchain, so the Apple Intelligence post-processing provider is compiled as a stub and does not perform live LLM cleanup in this artifact. The deterministic correction dictionary is unaffected.
• The F3 replace behavior (backspacing the previous text, then pasting) is new and best-effort; it works from wherever your cursor sits right after a dictation. Verify it in your editor of choice.