This update focuses on recording, transcription, and stability with large session libraries. Pin important sessions for quicker access Improved live transcription finalization to reduce unnecessary waiting Cloud summaries no longer wait for local post-processing Improved loading and scrolling performance for large session libraries Fixed a potential recording crash under memory pressure Fixed an issue where Quick Note text could occasionally become invisible Fixed untranslated Chinese text appearing in the English interface Improved background processing status and overall stability

Fix: tail-catch-up deadlock at .fair thermal state, which also blocked cloud summaries Fix: speaker-model warm-up gate was stricter than the run gate, so the model could never become ready Fix bug

This update focuses on transcription stability, terminology accuracy, and on-device performance: Added professional vocabulary packs for Technology, Healthcare, Legal, Finance, and Education, with your last selection remembered. Improved live transcription and fixed issues where new transcript segments could stop appearing or fail to recover after switching apps. Improved live speaker detection, speaker naming, and voiceprint matching. Optimized asynchronous model loading and thermal management, always prioritizing live transcription. Added optional speech enhancement for noisy environments. Cloud summaries no longer wait for local refinement or speaker processing to finish. Improved processing progress and status messages with fewer distracting notifications. Fixed various issues and improved overall reliability.

New Features 1.Upgraded live transcription engine (Chinese–English): a brand-new mixed zh-en recognition engine — real-time text as you speak, automatic language switching, and more accurate results. 2.Segment-level smart refinement: after recording, a stronger model re-checks the transcript to further improve accuracy. 3.Speaker diarization with speaker refinement: automatically tells speakers apart — shown live, then fully refined after recording for more accurate results — for 2–4 speaker conversations. 4.On-demand local AI model, smaller app: the on-device large model is no longer bundled with the app — it downloads on demand when needed, so the initial install is smaller and lighter. Improvements & Fixes 1.Speaker numbers now start at "Speaker 1" in order of appearance (no more starting mid-count). Speakers appear the moment recording ends, then refine in the background while numbers stay stable. 2.More reliable live transcription with clear status: during silence or brief load it shows "Live transcription is catching up — recording keeps saving," instead of freezing silently; recording never stops writing. 3.English recordings are no longer mis-refined by the mixed model, improving English accuracy. Various stability and performance improvements.

What's New 1.Fixed an occasional crash when starting Live Transcription. Previously, if a speech-model file became corrupted during the first install, the app could crash repeatedly and a restart wouldn't help — reinstalling was the only fix. The app now automatically detects and repairs damaged model files on launch, so a reinstall is no longer needed. 2.Improved the reliability of speech-engine loading.

What's New · Stability & polish 1.Fixed a rare crash when starting a recording while the mic was in use by another app or while switching Bluetooth/headphone routes — recording now starts more reliably. 2.Improved resilience during recording: if something goes wrong mid-session, your captured audio is now properly saved instead of being lost. 3.Optimized the transcription engine's resource handling for steadier, lower-memory performance on long sessions. 4.Cloud summaries now automatically retry when the provider is rate-limiting (e.g. too many requests) instead of failing outright. 5.Made the Ask chat faster to respond when you have many sessions — starting a question no longer hangs the UI. 6.Fixed a visual judder where a row would flicker when the delete confirmation appeared while swiping to delete a session.

Improved presets for weekly reports, todos, decisions, risks, and highlights. Better long-context handling in cloud mode to reduce missing information. Reuse previous answers for follow-up questions, making conversations faster and more consistent. Improved streaming output: scrolling up no longer forces you back to the bottom. Better Markdown rendering, related recording folding, keyboard handling, and scrolling. Fixed several stability and interaction issues.

1. Added recording-based Chat: ask questions about your existing recordings and quickly review meetings, projects, decisions, todos, and risks. 2. Added relevant recording retrieval: answers can include matched recording sources, making it easier to trace context. 3. Supports both on-device and cloud models: choose the mode that best fits your privacy, speed, and quality needs. 4. Improved Chat experience: streaming responses, Markdown rendering, saved conversation history, message list review, and smoother input and scrolling behavior. 5. Improved cloud model setup: after adding an API key, available models can be fetched and selected without manually typing model names. 6. Added first-run guided tips for Recording, Import, Session Detail, and Ask HushNote. 7. Fixed and refined multiple UI details, including Chat input positioning, related recording expansion, keyboard dismissal, and guided tip highlight alignment.

HushNote 1.1.2 Release Notes This update focuses on improving local summarization and long-recording workflows. Improved on-device summary model selection based on device memory. Optimized long-recording summarization for better stability and efficiency. Added resume support for summary processing, so interrupted local summaries can continue from completed progress when possible. Improved local model output quality, reducing repetitive, off-topic, or inaccurate results. Improved real-time transcription stability and reduced cases where catch-up transcription could pause recording. Added a subtle local model notice to help explain capability differences across devices and summary modes. Fixed several stability issues and polished the overall experience.

In this update: • Try the app for free before unlocking Pro • Unlock unlimited use with a one-time purchase • Quick notes can now be edited • Fixed bugs and improved overall stability

Support importing audio into HushNote from files and the Share Sheet Support on-device transcription processing for imported audio Support real-time progress display for imported audio tasks Support resumable imported audio tasks after interruptions Support additional transcription functions after transcription Support key-moment snapshots and quick notes during recording Add transcript refinement to improve cleanup quality for suspicious transcript segments Improve transcript refinement for Chinese and mixed Chinese-English spoken content Refine status feedback interactions in the recording view and imported audio detail view