Offline Speech to Text 0.7.0

Offline Speech to Text 0.7.0 makes long dictation finish sooner. While recording continues, OSTT now processes model-aware chunks in order. Parakeet and other conformer models use shorter sentence-sized work, while Whisper keeps longer runs that fit its encoder.

The speedup does not trade away the recording. OSTT retains the original audio until the final result and retries the exact complete message if any background chunk fails. The caller still receives one ordered final transcription.

Rapid app and editor changes are safer too. A stale recognition request releases capture ownership immediately, so the replacement request can start without a false no-audio or microphone-permission error while older native work winds down.

The compact APK contains no speech model. First-time setup recommends a verified model for the phone and downloads it only after the user chooses to proceed. Recognition stays on-device, and OSTT does not retain spoken audio or a transcription history.

The production-signed ARM64 APK was tested on a physical GrapheneOS Pixel with a 33-second dictation and an interrupt/restart stress run. Its SHA-256 is 87b2e148cf52cbee7797b3519eabb5f039ddf561ac95c9ebe298f9860f60c410.

Install from Zapstore, inspect the source, or download the verified APK.