Force regenerate applies to all selected pipeline steps

- Add --force flag to extract_caption.py: re-transcribe all videos
  even if caption files already exist
- Add --force flag to semantic_alignment.py: re-align all captions
  even if alignment files already exist
- Pipeline page now passes --force to transcribe and align steps
  (not just generate) when "Force regenerate" is checked
- Selecting only "Generate notes" + force → only regenerates notes
- Selecting "Transcribe" + "Align" + force → re-runs both stages

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: nodeeeeee

electron/renderer/app.js

@@ -1470,10 +1470,14 @@ async function attachPageHandlers() {
         chain.push(['Download videos', c]);
       }
       if (steps.includes('transcribe')) {
-        chain.push(['Transcribe', [python, paths.transcribe]]);
+        const c = [python, paths.transcribe];
+        if (force) c.push('--force');
+        chain.push(['Transcribe', c]);
       }
       if (steps.includes('align')) {
-        chain.push(['Align', [python, paths.align, '--course', cid]]);
+        const c = [python, paths.align, '--course', cid];
+        if (force) c.push('--force');
+        chain.push(['Align', c]);
       }
       if (steps.includes('generate')) {
         const c = [python, paths.generate, '--course', cid, '--course-name', name || courseNameFromId(cid), '--detail', detail, '--per-video'];

extract_caption.py

@@ -45,6 +45,7 @@
 # "api"   → always use OpenAI Whisper API
 WHISPER_BACKEND   = os.environ.get("AUTONOTE_WHISPER_BACKEND", "auto")
 VRAM_THRESHOLD_MIB = 15 * 1024   # ~15 GB nominal — accepts 16 GB cards (which report ~15.4 GB)
+FORCE_REGEN       = False        # set via --force CLI flag
 
 WHISPER_MODEL_SIZE = "large-v3"
 WHISPER_BEAM_SIZE  = 5
@@ -232,7 +233,7 @@ def transcribe_api(video_path: Path, caption_path: Path) -> bool:
     """
     import time as _time
 
-    if caption_path.exists():
+    if not FORCE_REGEN and caption_path.exists():
         print(f"  [skip] Caption already exists: {caption_path.name}")
         return True
 
@@ -417,7 +418,7 @@ def transcribe_local(video_path: Path, caption_path: Path) -> bool:
     Transcribe a video file directly with faster-whisper large-v3 on GPU.
     faster-whisper uses ffmpeg internally to decode audio from the video.
     """
-    if caption_path.exists():
+    if not FORCE_REGEN and caption_path.exists():
         print(f"  [skip] Caption already exists: {caption_path.name}")
         return True
 
@@ -580,7 +581,7 @@ def process_video(video_path: Path, manifest: dict, manifest_key: str | None) ->
     course_dir   = video_path.parent.parent     # [project]/[course_id]/
     caption_path = course_dir / "captions" / f"{video_path.stem}.json"
 
-    if caption_path.exists():
+    if not FORCE_REGEN and caption_path.exists():
         print(f"  [skip] Already captioned: {video_path.name}")
         if manifest_key and manifest_key in manifest:
             manifest[manifest_key]["caption"] = str(caption_path)
@@ -601,17 +602,24 @@ def process_video(video_path: Path, manifest: dict, manifest_key: str | None) ->
 # ── Entry point ───────────────────────────────────────────────────────────────
 
 def get_pending(manifest: dict) -> list[tuple[str, str]]:
-    """Return (key, video_path) for downloaded videos not yet captioned."""
+    """Return (key, video_path) for downloaded videos not yet captioned.
+
+    When FORCE_REGEN is True, returns ALL downloaded videos regardless of
+    whether captions already exist.
+    """
     pending = []
     for key, entry in manifest.items():
         if entry.get("status") != "done":
             continue
         vpath = entry.get("path")
         if not vpath or not Path(vpath).exists():
             continue
-        caption = Path(vpath).parent.parent / "captions" / f"{Path(vpath).stem}.json"
-        if not caption.exists():
+        if FORCE_REGEN:
             pending.append((key, vpath))
+        else:
+            caption = Path(vpath).parent.parent / "captions" / f"{Path(vpath).stem}.json"
+            if not caption.exists():
+                pending.append((key, vpath))
     return pending
 
 
@@ -620,8 +628,14 @@ def main() -> None:
     parser = argparse.ArgumentParser(description="Extract captions from Canvas lecture videos")
     parser.add_argument("--video", metavar="PATH",
                         help="Process a single video file (ignores manifest)")
+    parser.add_argument("--force", action="store_true",
+                        help="Re-transcribe all videos even if captions already exist")
     args = parser.parse_args()
 
+    global FORCE_REGEN
+    if args.force:
+        FORCE_REGEN = True
+
     manifest = load_manifest()
 
     if args.video:

semantic_alignment.py

@@ -1578,7 +1578,8 @@ def _load_mapping(mapping_path: Path, course_dir: Path) -> dict[str, list[Path]]
 
 
 def process_course(course_id: int | str, use_jina: bool = False,
-                   mapping_path: Path | None = None) -> None:
+                   mapping_path: Path | None = None,
+                   force: bool = False) -> None:
     course_dir  = COURSE_DATA_DIR / str(course_id)
     captions_dir = course_dir / "captions"
     print(f"Course dir : {course_dir}", flush=True)
@@ -1677,15 +1678,16 @@ def process_course(course_id: int | str, use_jina: bool = False,
                 continue
 
         # Check if all output files for this group already exist
-        if len(slide_group) == 1:
-            out_file = out_dir / f"{cap.stem}.json"   # legacy single-file naming
-            if out_file.exists():
-                print(f"  [skip] Already aligned: {cap.stem}")
-                continue
-        else:
-            if all((out_dir / f"{sp.stem}.json").exists() for sp in slide_group):
-                print(f"  [skip] Already aligned: {[sp.name for sp in slide_group]}")
-                continue
+        if not force:
+            if len(slide_group) == 1:
+                out_file = out_dir / f"{cap.stem}.json"   # legacy single-file naming
+                if out_file.exists():
+                    print(f"  [skip] Already aligned: {cap.stem}")
+                    continue
+            else:
+                if all((out_dir / f"{sp.stem}.json").exists() for sp in slide_group):
+                    print(f"  [skip] Already aligned: {[sp.name for sp in slide_group]}")
+                    continue
 
         # Use Jina multimodal for single PDF slides if available
         if use_jina and len(slide_group) == 1 and slide_group[0].suffix.lower() == ".pdf":
@@ -1726,6 +1728,8 @@ def main() -> None:
                         choices=["bge-m3", "jina", "mpnet"],
                         help="Embedding model for --suggest-matches "
                              "(default: bge-m3, alternatives: jina, mpnet)")
+    parser.add_argument("--force", action="store_true",
+                        help="Re-align all captions even if alignment files already exist")
     args = parser.parse_args()
 
     if args.suggest_matches:
@@ -1750,7 +1754,8 @@ def main() -> None:
 
     if args.course:
         mapping = Path(args.mapping) if args.mapping else None
-        process_course(args.course, use_jina=args.jina, mapping_path=mapping)
+        process_course(args.course, use_jina=args.jina, mapping_path=mapping,
+                       force=args.force)
         return
 
     if not args.caption or not args.slides: