FluidVoice v1.6.10-beta.8

What's New

• Added Fluid Speech Mini and Pico: smaller, English-only local models with updated weights and support for custom words and pronunciation training.
• Update compatible speech-model weights without updating the app. Downloads resume after interruption, verify files, and preserve the existing model if an update fails or is canceled.
• Pin transcripts beside their timestamp, keep them at the top of History, and switch between All and Pinned.
• Play audio from individual meeting transcript segments.
• Double-click a meeting paragraph to correct it, undo corrections, and include edits in copy/export. Automatically named meetings now include their recording date and time.
• Choose from Parakeet v3's 25 supported languages for final meeting transcripts. Live captions remain English-only.
• Added a local API to list meetings and retrieve transcripts as text or JSON.

Improvements

• Faster History search with lower memory use, reliable result selection, and a Retry action when search is unavailable.
• English onboarding recommends Fluid Speech Mini. Mini/Pico show English Only, and model Activate buttons are clearer.
• Quiet update notifications keep dictation responsive. Use the top-bar update icon or control update pop-ups in General settings.
• Settings backups include meeting defaults, model idle-unload timing, and per-prompt shortcuts and AI choices. Imports require active recording and model work to finish first.

Fixes

• Escape saves canceled dictation to History without inserting it, when History is enabled.
• Escape cancels pending voice Edit and Command actions, prevents late text insertion, and preserves completed results and rejected follow-up text.
• Fixed Control + Command capture, mouse/keyboard shortcut release handling, false triggers, and recording starts after dismissed dictionary suggestions.
• Switching from voice Edit to Dictation keeps the microphone recording.
• Updates verify the app before installation and preserve a recovery copy. Fixed stuck shutdowns and leftover update windows from an older running copy.
• Readable Command chats and file transcripts survive another saved history entry failing to decode.
• Fixed meeting memory growth after audio-write failures, improved interruption recovery and diagnostics, and cleared stale playback warnings.
• Fixed non-English meeting transcription using an outdated English-only fallback.
• Fixed dictionary spelling capture and preserved pronunciation recordings across speech-model versions, with retraining guidance for older recordings.
• Improved model download cleanup and safe deletion without changing unrelated selections; added repair guidance for missing Fluid Intelligence model files.
• Fixed GPT-6 verification/request compatibility and model reasoning appearing in cleaned-up dictation.
• Fixed the meeting More button shaking and missing clicks around its edges; improved dropdown click areas throughout the app.
• Edit and Command logs record instruction length instead of the spoken instruction text.
• Fixed repeated model-preview resizing, overlapping sidebar controls, lingering notch overlays, and dismissed microphone notices continuing to animate.
• Improved settings backup restore safety and prevented overlapping imports/exports.
• Improved Xcode 26.3 build compatibility.

Notes

• Meeting summaries are temporarily disabled while accuracy is improved.

Contributors

Thanks to @grohith327, @Lstarsky0, @nishantkumar1292, and @postoso for the 1.6.10-beta.8 fixes and improvements.