Other Changes
- (perf) Paste right away in apps that ignored an Accessibility insertion before (#1858)
- (bug) Read text fields in Chrome, which ignores AXManualAccessibility (#1856)
- (perf) Build the dictionary term prompt and hints once per load, and not at stop for a live result (#1847)
- (chore) bump the bundler-minor-patch group with 19 updates (#1853)
- (perf) Keep the Recorder's input levels and sub-second duration ticks out of its view model updates (#1840)
- (perf) Cache meeting audio process identities instead of re-reading them from CoreAudio every second (#1848)
- (perf) Skip indicator screen refreshes on clicks and app switches while no indicator is visible (#1841)
- (perf) Reuse a compiled meeting-tab script instead of spawning osascript every second (#1849)
- (perf) Stop rewriting history records without words on every launch (#1845)
- (perf) Find the recordings with stored voices that are still in History with one fetch at launch (#1844)
- (perf) Scan the installed apps off the main thread when the Workflows settings appear, not at launch (#1842)
- (perf) Reuse watch folder checksums of unchanged files instead of re-reading every file on each scan (#1827)
- (perf) Append API request chunks in place instead of copying the whole request per chunk (#1826)
- (chore) Update the README for stable 1.8.2 (#1852)
- (chore) Start pull request titles with a category prefix (#1851)
- (perf) Keep the LLM prompt vocabulary until the dictionary is loaded again (#1846)
- (perf) Stop comparing every dictionary entry on each keystroke in the entry editor (#1850)
- (perf) Stop rebuilding every dictionary row on each keystroke in the search field or entry editor (#1843)
- (perf) Apply dictionary corrections without re-reading every rule and re-bridging the text per rule (#1828)
- (perf) Normalize each word once when converting spoken numbers (#1825)
- (perf) Let only the indicator's dot and waveform observe the input level (#1832)
- (perf) Build the indicator presentation once per update instead of on every read (#1831)
- (perf) Publish the dictation recording duration only when the indicator would show a change (#1830)
- (perf) Keep the indicator views of inactive styles out of the SwiftUI graph (#1829)
- Free Granite TurboCTC memory after each transcription and on unload (#1836)
- Stop toggle dictations sooner when the speaker finished a thought, with Reson8's turn detection (#1833)
- Offer Rumble-2-Flash, live dictation, a dictionary custom model, pattern sets, filler words and speaker labels in Reson8 (1.1.0) (#1823)
- Move the Soniox plugin to STT v6 and refresh model lists saved before it (1.2.15) (#1824)