Skip to content

CHANGELOG.md

Latest commit

 

History

History
79 lines (62 loc) · 3.69 KB

CHANGELOG.md

File metadata and controls

79 lines (62 loc) · 3.69 KB

Changelog

[0.1.4] - 2026-03-17

Added

  • Numeric IDs in bilbo list output; all commands now accept a numeric ID instead of a title (e.g., bilbo info 1) (f9138b0)
  • Command aliases: remove → delete, run → process (f9138b0)
  • bilbo help command (89936d2)
  • Live VAD progress (EN VAD: 47%) during segmentation stage using Silero's progress_tracking_callback (bb266b8)

Changed

  • bilbo process audio arguments are now optional to allow re-running stages via --from/--to without re-specifying audio files (89936d2)
  • Removed per-stage CLI commands (bilbo transcribe, bilbo segment, bilbo align, bilbo export) in favour of bilbo process --from N --to N (89936d2)
  • Default Whisper model changed from large-v3-turbo to large-v3 (4ce3820)
  • Transcription now runs sequentially (L1 then L2) rather than in parallel; a single shared model cannot run concurrently (4ce3820)
  • refine_timestamps now uses ownership-based VAD matching: a speech region is assigned to the segment that contains >50% of its duration, and both .start and .end are snapped to the matched regions' extremes with ±50 ms padding (bb266b8)

Fixed

  • Assembly no longer redundantly reopens audio files for each pair (23c6729)

[0.1.3] - 2026-03-14

Added

  • Ported full Bertalign two-pass alignment algorithm (overlap encoding, anchor DP, m:n second pass with margin scoring) replacing the previous anchor+gap-fill approach (2bd8bc1)
  • GitHub Actions release workflow: auto-publish to PyPI on tag, TestPyPI .devN builds on main push (4c79154)
  • Log skipped sentences during alignment (98aab78)

Changed

  • Cover merging uses vectorized numpy instead of per-row loop (7118f92)
  • refine_timestamps reuses a single open SoundFile instead of re-reading per segment (7118f92)
  • find_problematic_regions uses numpy cumsum for sliding window (7118f92)
  • Deduplicated chapter-finishing logic in map_chapters_to_output (7118f92)
  • Fixed Library.rename to remap audio paths by directory component, not string replace (7118f92)

[0.1.2] - 2026-03-13

Added

  • Language auto-detection (omit --l1/--l2 to detect from audio) (6080564)
  • bilbo rename command to rename books (6080564)
  • bilbo transcribe, bilbo segment, bilbo align per-stage commands (6080564)
  • Slug-based filesystem paths (titles can now contain spaces and special characters) (6080564)

Changed

  • CLI now accepts human-readable titles instead of slugs (6080564)
  • Library index keyed by auto-generated slugs internally (6080564)

Fixed

  • --device auto now correctly resolves to CUDA/CPU (f9f2ec8)

[0.1.1] - 2026-03-12

Added

  • LLM-powered metadata merging (via local ollama) (86f3ff2)
  • Diagonal poster/cover merging (d8e80a0)
  • Warning tones around misaligned regions (d3d1ac3)
  • Better problem passage detection with sliding-window smoothing (d3d1ac3)
  • Energy-based segment timestamp extension (d64d16e)
  • PyPI release (pip install bilbo-audiobook) (8b0dfdf)

Changed

  • Improved logging and progress reporting (4afcf26)
  • Refactored CLI with Click (d64d16e)

Fixed

  • Metadata no longer lost on re-export (ab003a0)

[0.1.0] - 2026-03-10

Added

  • Initial release (dfce168)
  • 4-stage pipeline: transcription, segmentation, alignment, assembly (dfce168)
  • faster-whisper transcription with word-level timestamps (dfce168)
  • pySBD sentence segmentation (dfce168)
  • LaBSE cross-lingual alignment (anchor + gap-fill DP) (7475dbe)
  • Audio assembly with LUFS normalization, gaps, and crossfades (38f8bee)
  • Text export of aligned pairs (dd051fa)
  • Metadata extraction and cover art merging (a6ad026)
  • Multi-threaded parallel transcription (38f8bee)
  • Library management (bilbo list, bilbo info, bilbo delete) (dfce168)