Repository navigation
Releases: vrennat/hearsay
Release list
Hearsay 0.1.7
Live speaker labels now distinguish detected voices from speech whose speaker is uncertain.
- Unmatched words no longer silently become Remote 1 or inherit a nearby turn's speaker. Equal-overlap ambiguities also remain uncertain.
- Uncertain groups cannot be globally named or supply a name to the final transcript. Existing numbered voice labels and older meeting files remain compatible.
- Content-free diagnostics report segment slots and matched/uncertain word counts every 30 seconds, helping distinguish timing gaps from model merging.
- Publishing stages the signed app without replacing or restarting an installed copy.
This fixes unsupported speaker assignments and misleading labeling. It does not change the acoustic diarization model or guarantee that every participant will be separated. Final offline diarization remains independent.
Validation: 179 automated tests passed (142 XCTest and 37 Swift Testing), including synthetic four-voice, delayed-segment, ambiguous-overlap and final-name-transfer regressions. The real on-device title-generation test was excluded. No real call audio or transcripts were used. Live-call acoustic accuracy and UI interaction were not tested during the active recording.
Hearsay 0.1.6
What's Changed
- fix(release): inline update notes in the appcast by @vrennat in #6
- feat: move transcript lines between speakers; export decrypted audio by @vrennat in #10
- fix(meeting): inspector steps aside in narrow windows by @vrennat in #9
- fix(main): show the sidebar when opening with nothing selected by @vrennat in #11
- feat(diarizer): automatic call diarization by @vrennat in #12
Full Changelog: v0.1.5...v0.1.6
Hearsay 0.1.5
What's Changed
Full Changelog: v0.1.4...v0.1.5
Hearsay 0.1.4
What's Changed
Full Changelog: v0.1.3...v0.1.4
Hearsay 0.1.3
What's Changed
Full Changelog: v0.1.2...v0.1.3
Hearsay 0.1.2
Hearsay 0.1.1
Hearsay 0.1.0
First public release. Requires macOS 26 or later on Apple Silicon.