Skip to content

v2.5

Latest

Choose a tag to compare

@bisonbet bisonbet released this 07 Oct 01:56
2dd86b5

BisonNotes AI 2.5 (build 22) adds optional on-device transcript cleanup, brings maintenance tools together in Advanced Troubleshooting, and strengthens recording, restore, download, and iCloud deletion reliability.

App Store

Highlights

  • Readable English transcripts, on device. Optional cleanup uses S1-mini by Superwhisper to improve punctuation, remove fillers, and resolve spoken false starts on supported devices. It is off by default and requires an explicit model download.
  • Your original transcript stays available. Switch between Original and Cleaned; copy, share, and ordinary text export follow the selected version. Summaries continue to use the original text, and speaker identities and timing remain attached to the original segments.
  • One place for advanced maintenance. Advanced Troubleshooting consolidates storage and repair tools, reports scan results and detached rows, and protects active work during cleanup.
  • More durable iCloud deletions. Local deletions and their pending sync intents are persisted together, and deleted imported-audio placeholders no longer return on a later sync.
  • Safer recording and storage operations. Live transcription callbacks, restored-audio replacement, and model cache maintenance now handle their lifecycle and concurrency boundaries more reliably.

What's New

On-Device Transcript Cleanup

  • Automatic cleanup runs after a completed file transcription and any speaker labeling, including transcription reruns. It does not process live partial text or intermediate audio chunks.
  • A manual Clean up English transcript action also works on imported text transcripts without enabling automatic cleanup or rerunning transcription.
  • Cleaned text is stored alongside the original. Opening an existing transcript starts on Original; disabling cleanup or removing the model retains saved cleaned text.
  • The feature is English-only. Known non-English transcripts are skipped, and uncertain language produces an explanation rather than automatic processing.
  • Segment order, speaker mappings, and time ranges are preserved. Word-accurate timestamp exports continue to use the original transcript.
  • Cleanup validates output, bounds token use and runtime, and publishes a complete result only after every segment succeeds. Cancellation or failure keeps the original; completion is rejected if the source transcript changed or was deleted.
  • S1-mini has independent download, cancellation, removal, readiness, and bounded memory-cache handling. Its setting and model are local to each device.

Advanced Troubleshooting

  • Replaces the former Data Migration screen with consolidated Advanced Troubleshooting tools and service-layer maintenance logic.
  • Reports the actual orphaned-audio scan result and surfaces detached library rows and queued model deletions.
  • Revalidates reviewed items before deletion to close review-to-delete races and protects in-flight recording-combine output.
  • Improves handling of filenames containing colons and distinguishes retained sidecars from disposable files.
  • Aborts a partial clear-all operation rather than continuing through an incomplete deletion.

iCloud Sync and Storage

  • Persists deletion intent with the local deletion and carries earlier queued removals forward during migration.
  • Hardens imported-audio deletion markers and restore handling, including compatibility with legacy clients.
  • Prevents deleted imported placeholders from reappearing through iCloud.
  • Reports insufficient audio-staging space while allowing metadata sync to continue.
  • Coordinates cache maintenance with active model downloads and sync operations. Download-blob cleanup is deferred only for the model that owns the active download.
  • Restored audio is copied to a staging file before atomic replacement, preserving the previous destination if copying or installation fails.

Recording, Transcription, and Accessibility

  • Moves live audio and speech-recognition callbacks outside inherited main-actor isolation, and serializes audio-buffer work with stop/deactivation.
  • Releases the audio engine before asynchronous stop work can suspend.
  • Corrects transcription-rerun staleness checks and deleted-recording messaging, and propagates ASR language into background jobs.
  • Reports the underlying cause when a transcription cannot be saved.
  • Fixes the Record and Setup issues identified by the accessibility audits.

Documentation and Maintenance

  • Adds the v2.5 release and upgrade guide and updates the README and architecture documentation.
  • Removes dead files and retires superseded implementation plans.
  • Adds regression coverage for transcript cleanup, Advanced Troubleshooting, live transcription callbacks, restored audio, MLX downloads, and iCloud orchestration and backup behavior.

Notes

Transcript cleanup supports compatible physical iPhone/iPad devices and Apple-silicon native Mac builds; it is unavailable on Apple Watch and unsupported execution targets. Install updates over the existing app and let current recording, processing, and sync work finish first. Download S1-mini separately on each device where you want cleanup.

This release matches BisonNotes AI 2.5 (build 22), already built and released on the App Store.

Full Changelog: v2.4...v2.5