TypeWhisper 1.3.0-rc2 rolls up the stabilization work that landed after rc1, with a focus on push-to-talk reliability, short-speech handling, audio recovery hardening, and the refreshed local-model and licensing flows.
Reliability & Input Handling
- Fixed modifier-combo push-to-talk so transient
flagsChangedchurn does not stop recording before the final key release. - Added the default push-to-talk discard behavior when extra keys interrupt an active hold, while keeping the stop point predictable on the actual hotkey release.
- Added an advanced opt-in to transcribe more short and quiet clips instead of classifying them as no speech.
- Hardened audio recovery, preview recovery, and recording-abort cleanup around route changes and device loss.
- Delayed media playback resume after dictation to avoid premature restarts.
- Ignored benign startup configuration changes that should not trigger unnecessary recovery work.
Local Models & Plugins
- Upgraded the shared local-model runtime stack and added Gemma 4 support with clearer model-policy messaging.
- Improved local-model download, cancellation, and Hugging Face token flows across the bundled local plugins.
- Fixed Gemini model discovery and prompt-processing fallback handling.
Settings & UX
- Redesigned the in-app licensing settings flow.
- Aligned spoken-language picker labels and search behavior with the app locale.