Repository navigation
VoxKey 0.7.1
Live-preview improvements
- Reduce repeated preview decoding during longer phrases while preserving the
initial preview timing and the full preview context. - Catch previews up after a pause, including short utterances and pauses with
quiet background sound. Catch-ups survive busy decoding and rapidly resumed
speech, use the newest audio, and remain rate-limited. - Keep preview failures isolated from capture and final recognition. Preview
words remain presentation-only; only final transcription reaches the editor.
Reliability fixes
- Find focused editors in deeply nested windows without rejecting them because
of unrelated nested content. Node and time limits, cycle detection, ambiguous
focus rejection, and secure-field protection remain in place. - Fix audio-resampler compilation with newer Swift toolchains while retaining
the shared, synchronized converter callback state.
Privacy
Production builds contain no VoxKey telemetry, performance recording, debug
logging, transcription tracing, or diagnostic controls. These facilities require
explicit local-only compile flags and are excluded from the shipped executable.
Saved preferences cannot enable them in production.
Validation limits
Automated coverage includes preview scheduling, pause/resume behavior, busy
decoding, final-audio preservation, focused-editor resolution, and delivery safety.
Model checks use synthetic speech; they do not establish battery impact, accuracy,
or latency across all Macs and editors. The maintainer reports successful recent
daily use, but a fresh-user installation, physical microphone compatibility
matrix, VoiceOver walkthrough, and installed Sparkle update have not been repeated
for this release. The previously reported intermittent Right Command activation
failure remains unresolved.