Index transcripts into library search (FTS), both UIs

The 🔍 library search now matches spoken words, not just titles /
channels / descriptions.

- database.rs: add a `transcript` column to the `video_search` FTS5 index.
  Fresh DBs get it directly; existing ones are migrated (FTS5 has no ADD
  COLUMN, so the table is recreated and search_meta cleared to force a
  one-time reindex). `sync_search_index` reads each video's first subtitle
  (mtime-gated like the description), flattens it via the shared `vtt`
  parser (`transcript_text`, capped at 256 KB), and indexes it. Search
  snippets now use column -1 so the excerpt comes from whichever column
  matched (description or transcript).
- library.rs: `build_search_entries` passes the first subtitle path.
- Both UIs: search copy updated to mention transcripts.

Tests: a unit test (a word only in the .vtt is found), a migration test
(seed an old 5-column index + stale meta, open, confirm the recreated
index indexes transcript text), and the search integration test now seeds
a subtitle and asserts a spoken-only word hits. Also bumped the test
server's startup budget so the ffmpeg-heavy dedup test can't starve a
sibling server into a flaky timeout. 120 tests pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

src/app.rs

@@ -994,7 +994,7 @@ impl App {
             .show(ctx, |ui| {
                 let resp = ui.add(
                     egui::TextEdit::singleline(&mut self.search_query)
-                        .hint_text("titles, channels, descriptions…")
+                        .hint_text("titles, channels, descriptions, transcripts…")
                         .desired_width(f32::INFINITY),
                 );
                 if std::mem::take(&mut self.search_focus) {
@@ -1007,7 +1007,7 @@ impl App {
                 }
                 ui.separator();
                 if self.search_query.trim().is_empty() {
-                    ui.weak("Type to search every title, channel, and description in the library.");
+                    ui.weak("Type to search every title, channel, description, and transcript in the library.");
                 } else if self.search_results.is_empty() {
                     ui.weak("No matches.");
                 } else {
@@ -1377,7 +1377,7 @@ impl App {
                 );
                 ui.separator();
                 if ui.button("🔎 Search")
-                    .on_hover_text("Full-text search titles + descriptions across the whole library")
+                    .on_hover_text("Full-text search titles, descriptions + transcripts across the whole library")
                     .clicked()
                 {
                     self.show_search = true;

src/database.rs

@@ -46,6 +46,9 @@ pub struct SearchEntry {
     pub channel: String,
     pub title: String,
     pub description_path: Option<std::path::PathBuf>,
+    /// First subtitle file, if any. Its parsed text is indexed so search
+    /// matches spoken words. Read lazily, only for new/changed videos.
+    pub subtitle_path: Option<std::path::PathBuf>,
 }
 
 /// A full-text search result row from [`Database::search_videos`].
@@ -68,6 +71,20 @@ pub struct StoredFingerprint {
     pub hashes: Vec<u64>,
 }
 
+/// Flatten a subtitle file's cues into one searchable text blob, capped so a
+/// very long video can't bloat the index unboundedly. Uses the shared
+/// VTT/SRT parser, so styling tags and duplicate rolling lines are stripped.
+fn transcript_text(vtt: &str) -> String {
+    const CAP: usize = 256 * 1024;
+    let mut out = String::new();
+    for cue in crate::vtt::parse(vtt) {
+        if out.len() + cue.text.len() + 1 > CAP { break; }
+        if !out.is_empty() { out.push(' '); }
+        out.push_str(&cue.text);
+    }
+    out
+}
+
 /// Build a safe FTS5 MATCH expression from free-form user input: each
 /// whitespace token becomes a quoted prefix term, AND-ed together. Quoting
 /// neutralizes FTS5 operators in the input; the trailing `*` gives
@@ -238,15 +255,17 @@ impl Database {
             -- Full-text search over the library. `video_search` is a standalone
             -- FTS5 index (available in rusqlite's bundled SQLite); `search_meta`
             -- tracks each indexed video's mtime so [`Database::sync_search_index`]
-            -- only re-reads a description sidecar when the video actually
+            -- only re-reads a video's sidecars when it actually changed.
-            -- changed. video_id/platform are UNINDEXED — stored for retrieval,
+            -- video_id/platform are UNINDEXED — stored for retrieval, not
-            -- not matched.
+            -- matched. `transcript` holds the video's subtitle text so search
+            -- matches spoken words too.
             CREATE VIRTUAL TABLE IF NOT EXISTS video_search USING fts5(
                 video_id UNINDEXED,
                 platform UNINDEXED,
                 channel,
                 title,
                 description,
+                transcript,
                 tokenize = 'porter unicode61'
             );
             CREATE TABLE IF NOT EXISTS search_meta (
@@ -278,6 +297,28 @@ impl Database {
                 if msg.contains("duplicate column") => {}
             Err(e) => return Err(e),
         }
+
+        // ── Migration: video_search.transcript column ────────────────────
+        // FTS5 has no ADD COLUMN, so when the index predates the transcript
+        // column we recreate the table and clear search_meta to force a
+        // one-time reindex (which pulls in subtitle text). The index is a
+        // derived cache, so dropping it loses nothing permanent.
+        let has_transcript = {
+            let mut stmt = conn.prepare("PRAGMA table_info(video_search)")?;
+            let cols = stmt.query_map([], |r| r.get::<_, String>(1))?;
+            let mut found = false;
+            for c in cols { if c? == "transcript" { found = true; } }
+            found
+        };
+        if !has_transcript {
+            conn.execute_batch(
+                "DROP TABLE IF EXISTS video_search;
+                 CREATE VIRTUAL TABLE video_search USING fts5(
+                     video_id UNINDEXED, platform UNINDEXED, channel, title,
+                     description, transcript, tokenize = 'porter unicode61');
+                 DELETE FROM search_meta;",
+            )?;
+        }
         Ok(())
     }
 
@@ -993,15 +1034,19 @@ impl Database {
             if existing.get(&e.video_id) == Some(&e.mtime_unix) {
                 continue; // unchanged — leave the indexed row in place
             }
-            // Only new/changed videos pay the description-read cost.
+            // Only new/changed videos pay the file-read cost.
             let description = e.description_path.as_ref()
                 .and_then(|p| std::fs::read_to_string(p).ok())
                 .unwrap_or_default();
+            let transcript = e.subtitle_path.as_ref()
+                .and_then(|p| std::fs::read_to_string(p).ok())
+                .map(|vtt| transcript_text(&vtt))
+                .unwrap_or_default();
             tx.execute("DELETE FROM video_search WHERE video_id = ?1", [&e.video_id])?;
             tx.execute(
-                "INSERT INTO video_search (video_id, platform, channel, title, description)
+                "INSERT INTO video_search (video_id, platform, channel, title, description, transcript)
-                 VALUES (?1, ?2, ?3, ?4, ?5)",
+                 VALUES (?1, ?2, ?3, ?4, ?5, ?6)",
-                rusqlite::params![e.video_id, e.platform, e.channel, e.title, description],
+                rusqlite::params![e.video_id, e.platform, e.channel, e.title, description, transcript],
             )?;
             tx.execute(
                 "INSERT OR REPLACE INTO search_meta (video_id, mtime_unix) VALUES (?1, ?2)",
@@ -1033,10 +1078,12 @@ impl Database {
         let conn = self.conn();
         let mut stmt = conn.prepare(
             // STX/ETX (\u{2}/\u{3}) delimit matched terms — control chars
-            // that won't collide with literal '[' / ']' in a description.
+            // that won't collide with literal '[' / ']' in the text. Column
-            // The UI turns them into highlight markup.
+            // -1 lets FTS5 pull the snippet from whichever column matched
+            // (description or transcript). The UI turns the markers into
+            // highlight markup.
             "SELECT video_id, platform, channel, title,
-                    snippet(video_search, 4, char(2), char(3), '…', 12)
+                    snippet(video_search, -1, char(2), char(3), '…', 12)
              FROM video_search
              WHERE video_search MATCH ?1
              ORDER BY rank
@@ -1145,7 +1192,7 @@ mod search_tests {
         SearchEntry {
             video_id: id.into(), mtime_unix: mtime,
             platform: "channels".into(), channel: channel.into(),
-            title: title.into(), description_path: None,
+            title: title.into(), description_path: None, subtitle_path: None,
         }
     }
 
@@ -1194,6 +1241,64 @@ mod search_tests {
         assert!(db.search_videos("", 10).unwrap().is_empty());
         assert!(db.search_videos("   \"  ", 10).unwrap().is_empty());
     }
+
+    #[test]
+    fn migrates_pre_transcript_index() {
+        // Seed a file DB with the OLD 5-column FTS table (no transcript) and a
+        // stale search_meta row, then open it — init_schema must recreate the
+        // table with a transcript column and clear search_meta.
+        let path = std::env::temp_dir().join(format!("ytoff-mig-{}.db", std::process::id()));
+        let _ = std::fs::remove_file(&path);
+        {
+            let conn = Connection::open(&path).unwrap();
+            conn.execute_batch(
+                "CREATE VIRTUAL TABLE video_search USING fts5(
+                    video_id UNINDEXED, platform UNINDEXED, channel, title, description,
+                    tokenize='porter unicode61');
+                 CREATE TABLE search_meta(video_id TEXT PRIMARY KEY, mtime_unix INTEGER NOT NULL);
+                 INSERT INTO search_meta VALUES('stale', 1);",
+            ).unwrap();
+        }
+        let db = Database::open(&path).unwrap(); // runs the migration
+
+        // If the table were still 5-column, the 6-column INSERT in
+        // sync_search_index would error — so a transcript hit proves migration.
+        let dir = std::env::temp_dir().join(format!("ytoff-mig-vtt-{}", std::process::id()));
+        std::fs::create_dir_all(&dir).unwrap();
+        let vtt = dir.join("s.vtt");
+        std::fs::write(&vtt, "WEBVTT\n\n00:00:01.000 --> 00:00:02.000\nmagnificent aardvark\n").unwrap();
+        let e = SearchEntry {
+            video_id: "v".into(), mtime_unix: 5, platform: "channels".into(),
+            channel: "c".into(), title: "t".into(), description_path: None,
+            subtitle_path: Some(vtt),
+        };
+        db.sync_search_index(&[e]).unwrap();
+        assert_eq!(db.search_videos("aardvark", 10).unwrap().len(), 1);
+
+        let _ = std::fs::remove_file(&path);
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+
+    #[test]
+    fn indexes_transcript_text() {
+        let db = Database::open_in_memory().unwrap();
+        let dir = std::env::temp_dir().join(format!("ytoff-tr-{}", std::process::id()));
+        std::fs::create_dir_all(&dir).unwrap();
+        let vtt = dir.join("a.en.vtt");
+        std::fs::write(&vtt, "WEBVTT\n\n00:00:01.000 --> 00:00:03.000\nthe quick brown fox\n").unwrap();
+        let e = SearchEntry {
+            video_id: "a".into(), mtime_unix: 1, platform: "channels".into(),
+            channel: "Chan".into(), title: "Unrelated Title".into(),
+            description_path: None, subtitle_path: Some(vtt),
+        };
+        assert_eq!(db.sync_search_index(&[e]).unwrap(), 1);
+        // A word only spoken in the transcript still matches.
+        assert_eq!(db.search_videos("brown", 10).unwrap().len(), 1);
+        assert_eq!(db.search_videos("fox", 10).unwrap()[0].video_id, "a");
+        // A word in nothing misses.
+        assert!(db.search_videos("zebra", 10).unwrap().is_empty());
+        let _ = std::fs::remove_dir_all(&dir);
+    }
 }
 
 #[cfg(test)]

src/library.rs

@@ -702,6 +702,7 @@ pub fn build_search_entries(lib: &[Channel]) -> Vec<crate::database::SearchEntry
                 channel: ch.name.clone(),
                 title: v.title.clone(),
                 description_path: v.description_path.clone(),
+                subtitle_path: v.subtitles.first().map(|s| s.path.clone()),
             });
         }
     }

src/web_ui/index.html

@@ -205,7 +205,7 @@
     <option value="size-desc">Largest</option>
   </select>
   <span id="hdr-stats"></span>
-  <button onclick="openSearch()" title="Search titles + descriptions (f)">🔍</button>
+  <button onclick="openSearch()" title="Search titles, descriptions & transcripts (f)">🔍</button>
   <button onclick="shufflePlay()" title="Play a random unwatched video">🎲</button>
   <button onclick="rescan()" title="Rescan library">⟳</button>
   <button id="dl-btn" onclick="openDownloads()" title="Downloads">⬇<span id="dl-badge">0</span></button>
@@ -1516,7 +1516,7 @@ function openSearch(){
   bg.onclick=e=>{if(e.target===bg)bg.remove()};
   bg.innerHTML=`<div class="modal" style="max-width:720px;width:100%">
     <div class="modal-hdr"><h2>🔍 Search library</h2><button onclick="this.closest('.modal-bg').remove()">✕</button></div>
-    <input id="fts-input" type="search" placeholder="Search titles, channels, descriptions…" autocomplete="off"
+    <input id="fts-input" type="search" placeholder="Search titles, channels, descriptions, transcripts…" autocomplete="off"
       style="width:100%;padding:8px 10px;background:var(--bg);border:1px solid var(--border);border-radius:4px;color:var(--text);font-size:14px">
     <div id="fts-status" style="font-size:11px;color:var(--muted);padding:6px 2px"></div>
     <div id="fts-results" style="overflow-y:auto;max-height:64vh"></div>
@@ -2495,7 +2495,7 @@ function showShortcutsHelp(){
     <div class="modal-hdr"><h2>Keyboard shortcuts</h2><button onclick="this.closest('.modal-bg').remove()">✕</button></div>
     <table><tbody>
       ${row('/','Focus filter box')}
-      ${row('f','Search titles + descriptions')}
+      ${row('f','Search titles, descriptions & transcripts')}
       ${row('r','Rescan library')}
       ${row('d','Open downloads')}
       ${row('Esc','Close modal / cancel')}

tests/api.rs

@@ -66,7 +66,9 @@ impl Server {
     }
 
     fn wait_ready(&self) {
-        for _ in 0..150 {
+        // Generous budget: the ffmpeg-heavy dedup test can spike CPU and delay
+        // a sibling server's startup, so don't flake under parallel load.
+        for _ in 0..400 {
             if let Some((code, _)) = curl(&self.req_args("/", "GET"), None) {
                 if code != 0 { return; } // 0 = connection refused (not up yet)
             }
@@ -338,6 +340,9 @@ fn full_text_search_indexes_titles_and_descriptions() {
     std::fs::write(chan.join("Cooking Sourdough [vid123].mkv"), b"").unwrap();
     std::fs::write(chan.join("Cooking Sourdough [vid123].description"),
                    b"an artisan bread tutorial").unwrap();
+    // A subtitle sidecar whose spoken words appear nowhere in the title/desc.
+    std::fs::write(chan.join("Cooking Sourdough [vid123].en.vtt"),
+                   b"WEBVTT\n\n00:00:01.000 --> 00:00:04.000\nwhisk the poolish gently\n").unwrap();
 
     // Rescan so the server picks up the new file and reindexes it.
     let (code, _) = s.post("/api/rescan", "");
@@ -354,6 +359,9 @@ fn full_text_search_indexes_titles_and_descriptions() {
     // A word only in the description hits (proves the sidecar got indexed).
     assert!(s.get("/api/search?q=artisan").1.contains("vid123"), "description search should hit");
 
+    // A word only spoken in the transcript hits (proves the .vtt got indexed).
+    assert!(s.get("/api/search?q=poolish").1.contains("vid123"), "transcript search should hit");
+
     // An unrelated query does not.
     assert!(!s.get("/api/search?q=quantumchromodynamics").1.contains("vid123"),
             "unrelated query must not hit");