Skip to content

FlowLocal 1.2.11

Choose a tag to compare

View all tags
@kpsr01 kpsr01 released this 25 Sep 16:00
· 1 commit to main since this release
v1.2.11
17753b0

FlowLocal 1.2.11 switches to local, speaker-conditioned dictation.

  • Replaces Nemotron with Sortformer v2.1 diarization and INT8 Multitalker Parakeet streaming via parakeet-rs.
  • Adds local voice enrollment and clean-speech speaker matching; only the matched speaker's raw transcript is inserted.
  • Removes the language-model transcript cleanup and the previous model downloads. Existing history remains readable.
  • Downloads the new models on first launch. Open Settings → System status → Your voice to enroll before dictating.

English-only, CPU inference. Voice matching is best-effort, not security-grade biometric authentication. Short or overlapping speech may produce no insertion; see Known Limitations in the repository.

Validation: solution build and native unit test passed; mixed-voice smoke recognized the enrolled voice and rejected a different voice alone. Of 206 .NET tests, 203 passed, two skipped, and one foreground-window smoke test failed at SetForegroundWindow in this desktop session.

Assets 4
Loading