FluidVoice v1.6.13

What's New

• Fluid 1.1 Ripple and Splash: updated on-device cleanup, with Ripple offering a faster alternative to Splash. Both are research previews; review your text before sending.
• Fluid Speech Pro: our most accurate local dictation model for Apple Silicon. Pro, Mini and Pico support 25 languages.
• Edit Mode (Beta): a dedicated Edit 1.1 model on Apple Silicon, with settings separate from dictation and continued support for your own API providers.
• Qwen3 ASR 1.7B: on-device dictation on Apple Silicon with automatic detection of 30 languages.
• Apple Intelligence for Command Mode: use Apple's on-device model on supported Macs running macOS 27 or later.

Improvements

• Compact, consistent start and end prompts with visible countdowns and clearer button feedback.
• Detected calls stop recording after a cancellable countdown. Muting or losing a call window no longer ends an active recording prematurely.
• Automatic captures computer audio and your microphone, including when recording again. Online meetings no longer silently switch to microphone-only recording.
• Choose the meeting app and microphone before recording, with recent choices first and better handling of disconnected microphones and restarted apps.
• Rename meetings while recording. Renaming a saved meeting keeps its summary.
• Meeting transcripts use Fluid Speech Pro with an available Parakeet fallback. Redo a finished transcript in another supported language using its saved audio.
• Saved dictionary replacements now apply to completed meeting transcripts. Redo also refreshes replacements, with improved English number and currency formatting.
• Smoother live captions with Reduce Motion support. Local tools can also read live captions through the local API.
• More consistent formatting for spoken quantities and currencies, plus named emoji, while preserving code, filenames and identifiers.
• Accepted dictionary corrections save their audio and original/corrected text locally for future training. Nothing is uploaded, and FluidVoice tells you if audio cannot be saved.
• Choose how soon dictionary suggestions appear, without repeated delays while editing.
• Dashboard recommends speech and cleanup models for your Mac, with download-and-activate actions and dismissible suggestions.
• AI Providers is now AI Settings, with clearer Edit Mode and Cleanup Styles controls. Restoring older settings preserves your Edit provider.
• Choose an Apple Speech language and region on macOS 26 or later.
• Choose your own accent color in Settings.
• New recording sounds and stop cues, with sound settings now under Dictation.
• Shortcuts now distinguish left and right modifier keys. Re-record existing shortcuts to choose a specific side.
• Fluid Intelligence screen context is off by default, with optional setup and a reminder to review AI changes.

Fixes

• Preserve opening speech, short replies and sentence endings, with more readable overlapping speech and fewer duplicate words.
• Reduce false incomplete-transcription warnings during normal recording startup and shutdown.
• Keep previous transcripts when redo fails or is interrupted, and keep loaded meeting history visible if saved files cannot be read.
• Improved Google Meet detection and more reliable saved recording preferences.
• Trained words keep working after reloading Fluid Speech Pro.
• Better first-word recognition in short Parakeet dictations, with fixes for skipped short audio and final snippets.
• More reliable pasting and recovery in slow editors, web editors and consecutive dictations. Undelivered text stays on the clipboard for manual recovery.
• Improved text-field detection in Electron apps and restored dictation in Muse.
• Verify, update or delete downloaded cleanup models without activating them first. The active model stays clearly marked.
• More reliable model downloads, automatic retries for interrupted speech downloads, and uninterrupted setup when resizing Dashboard.
• Better alignment and readability in the Small dictation overlay.
• Clearer Cancel Recording shortcut validation. Unsupported modifier-only assignments reset to Esc; valid shortcuts and disabled settings are preserved.
• Preserve older dictation history and its saved audio when updating FluidVoice.
• Turning off Read my screen discards captured context. Screen-word matching handles everyday contractions more accurately.
• Fix simultaneous transcript exports mixing content or overwriting temporary files.
• Stronger local API protection against browser-based access and clearer meeting-sharing permission errors.

Contributors

Thanks to @ParamKataria, @grohith327, @TokuiNico, @sabbah13, @postoso, @andrewmitchell-automations, @Livvux and @BenjaminPrice for 1.6.13 fixes and improvements.