feat(stats): add alass sidecar retiming for sentence mining and fix timi

- Retime local English sidecars against the Japanese sidecar via alass before populating sentence card translation fields; cache retimed copies for the process lifetime - Reject reversed or non-positive subtitle timings in immersion tracker and before media generation - Fix word card mining so sentence audio goes to SentenceAudio (not ExpressionAudio) and English subtitle text is not written to SelectionText - Consolidate 10 individual change fragments into stats-updates.md
2026-07-30 07:21:32 -07:00 · 2026-06-06 14:26:02 -07:00
parent 01c2d9eb1d
commit 99401e5a70
31 changed files with 985 additions and 123 deletions
@@ -1,4 +0,0 @@
-type: fixed
-area: stats
-
- Retried and batched stats home cover images from stored DB art so AniList art appears without extra AniList lookups or a full page refresh.
@@ -1,9 +0,0 @@
-type: fixed
-area: stats
-
- Fixed vocab-page example sentence mining buttons failing when the Anki deck setting is blank or Yomitan card formats are ordered with a non-term card first.
- Fixed vocab-page example word and audio mining so English subtitles are only written to Selection Text for sentence cards, leaving word cards to show the normal Yomitan dictionary glossary.
- Fixed stats-page sentence mining audio updates so generated sentence clips populate `SentenceAudio` without also writing the same clip to the configured expression-audio field.
- Fixed stats-page word mining so the hidden Yomitan helper uses the same configured word-audio sources as the normal Yomitan plus button.
- Fixed stats-page sentence mining to use the current Anki deck/settings at request time, create direct sentence cards before slow media generation completes, and include stored English subtitle text as Selection Text.
- Fixed secondary subtitle auto-selection to prefer regular English tracks over Signs/Songs tracks when both are available.
@@ -1,4 +0,0 @@
-type: fixed
-area: stats
-
- Fixed vocabulary exclusions so adding a word once hides matching token variants across the vocabulary page and duplicate exclusions are collapsed.
@@ -1,4 +0,0 @@
-type: changed
-area: stats
-
- Renamed the vocabulary detail “Similar Words” section to “Related Seen Words” and tightened matches to same readings or shared kanji, avoiding noisy kana-suffix matches.
@@ -1,4 +0,0 @@
-type: added
-area: stats
-
- Showed a progress indicator while sessions are being deleted on the home and sessions pages, so long batch deletes give visible feedback instead of appearing to do nothing.
@@ -1,4 +0,0 @@
-type: added
-area: stats
-
- Added a Hide Kana filter to the common-words frequency table so kana-only headwords can be hidden while reviewing mining targets.
@@ -1,4 +0,0 @@
-type: fixed
-area: stats
-
- Library card size selection is now remembered across Stats window reloads and remounts.
@@ -1,5 +0,0 @@
-type: added
-area: stats
-
- Added headword-based sentence search by default, so searches like `知らない` can find tracked lines containing inflected variants such as `知らねえ` while exact sentence text searches still work.
- Added a Search by headword toggle on the Stats search page for switching back to exact text/title matching.
@@ -1,6 +0,0 @@
-type: added
-area: stats
-
- Added a Search tab for realtime subtitle sentence search with media context and mining actions.
- Search results can mine sentence cards from valid source lines, while word/audio card actions appear only for exact searched-word matches.
- Search results omit secondary subtitle text from display and matching, but pass stored secondary subtitle text into sentence-card mining when available.
@@ -1,4 +0,0 @@
-type: fixed
-area: stats
-
- Sped up deleting stats sessions by refreshing only affected rollups and rebuilding lifetime summaries with aggregate SQL.
@@ -0,0 +1,8 @@
+type: changed
+area: stats
+
+- Added the Stats Search tab for realtime subtitle sentence search with media context, headword matching, and mining actions for source-backed sentence cards or exact-match word/audio cards.
+- Improved Stats mining from Search and vocabulary examples: empty `ankiConnect.deck` can use Yomitan's mining deck, sentence cards are created before slow media generation finishes, secondary subtitles from stored lines, sidecar files, or temporary alass-retimed English sidecars populate sentence Selection Text, invalid stored timings are blocked before FFmpeg runs, future out-of-order subtitle timing pairs are skipped until valid timings arrive, and partial media failures are shown.
+- Fixed Stats mining field/audio behavior so sentence clips update `SentenceAudio`, word audio uses the configured Yomitan sources, English subtitle text is not written onto word cards, and secondary subtitle auto-selection prefers regular English tracks over Signs/Songs tracks.
+- Improved vocabulary review with a Hide Kana filter, duplicate-collapsed exclusions across token variants, and Related Seen Words matching based on shared readings or kanji.
+- Improved Stats browsing reliability by remembering library card size, retrying stored cover art without extra AniList lookups, showing progress during session deletes, and making session deletes refresh faster.