Add transcription and alignment pipeline

New commands:
- `transcribe`: Uses Whisper to generate word-level timestamps from video
- `align`: Matches manuscript markers to transcript, outputs transcript.csv

Workflow:
1. gnommo transcribe video.mov → video.transcript.json
2. gnommo align project/ → transcript.csv with markers at aligned times

Alignment uses fuzzy text matching to find the first phrase after each
marker in the manuscript, then locates it in the transcript. Applies
configurable offset (default -1s) so slides appear before speech.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

gnommo/cli.py

@@ -18,6 +18,8 @@ from .parser import (
 from .validator import validate_project
 from .transformer import build_render_plan
 from .renderer import render, generate_ffmpeg_command_string
+from .transcriber import transcribe_video, save_transcript, load_transcript
+from .aligner import align_markers, save_aligned_transcript
 
 
 def main() -> int:
@@ -87,6 +89,50 @@ def main() -> int:
         help="Slide type for all slides (default: square)",
     )
 
+    # transcribe command
+    transcribe_parser = subparsers.add_parser(
+        "transcribe",
+        help="Transcribe video audio using Whisper",
+    )
+    transcribe_parser.add_argument(
+        "video",
+        type=Path,
+        help="Path to video file",
+    )
+    transcribe_parser.add_argument(
+        "-o", "--output",
+        type=Path,
+        help="Output JSON file (default: <video>.transcript.json)",
+    )
+    transcribe_parser.add_argument(
+        "--model",
+        default="base",
+        choices=["tiny", "base", "small", "medium", "large"],
+        help="Whisper model size (default: base)",
+    )
+
+    # align command
+    align_parser = subparsers.add_parser(
+        "align",
+        help="Align manuscript markers to transcript timestamps",
+    )
+    align_parser.add_argument(
+        "project",
+        type=Path,
+        help="Path to project directory",
+    )
+    align_parser.add_argument(
+        "--transcript",
+        type=Path,
+        help="Path to transcript JSON (default: media/talking_head.transcript.json)",
+    )
+    align_parser.add_argument(
+        "--offset",
+        type=float,
+        default=-1.0,
+        help="Seconds to offset marker times (default: -1.0)",
+    )
+
     args = parser.parse_args()
 
     try:
@@ -97,6 +143,11 @@ def main() -> int:
             return cmd_render(args.project, output, args.verbose, args.dry_run)
         elif args.command == "generate-slides":
             return cmd_generate_slides(args.directory, args.type)
+        elif args.command == "transcribe":
+            output = args.output or args.video.with_suffix(".transcript.json")
+            return cmd_transcribe(args.video, output, args.model)
+        elif args.command == "align":
+            return cmd_align(args.project, args.transcript, args.offset)
     except GnommoError as e:
         print(f"Error: {e}", file=sys.stderr)
         return 1
@@ -231,5 +282,81 @@ def cmd_generate_slides(directory: Path, slide_type: str) -> int:
     return 0
 
 
+def cmd_transcribe(video_path: Path, output_path: Path, model: str) -> int:
+    """Transcribe video audio using Whisper."""
+    print(f"Transcribing: {video_path}")
+    print(f"Model: {model}")
+    print()
+
+    words = transcribe_video(video_path, model=model)
+
+    print(f"  - Transcribed {len(words)} words")
+    print(f"  - Duration: {words[-1].end:.1f}s" if words else "  - No words found")
+
+    save_transcript(words, output_path)
+    print(f"  - Saved to: {output_path}")
+
+    # Show first few words as preview
+    if words:
+        preview = " ".join(w.word for w in words[:10])
+        print(f"  - Preview: {preview}...")
+
+    return 0
+
+
+def cmd_align(project_path: Path, transcript_path: Path = None, offset: float = -1.0) -> int:
+    """Align manuscript markers to transcript timestamps."""
+    print(f"Aligning: {project_path}")
+    print(f"Offset: {offset}s")
+    print()
+
+    # Load manuscript
+    manuscript_path = project_path / "manuscript.txt"
+    if not manuscript_path.exists():
+        print(f"Error: manuscript.txt not found", file=sys.stderr)
+        return 1
+
+    manuscript_text = manuscript_path.read_text(encoding="utf-8")
+
+    # Load transcript
+    if transcript_path is None:
+        # Try to find transcript in media folder
+        transcript_path = project_path / "media" / "talking_head.transcript.json"
+
+    if not transcript_path.exists():
+        print(f"Error: Transcript not found: {transcript_path}", file=sys.stderr)
+        print("Run 'gnommo transcribe' first to generate the transcript.", file=sys.stderr)
+        return 1
+
+    print(f"  - Loading transcript: {transcript_path}")
+    transcript = load_transcript(transcript_path)
+    print(f"  - Loaded {len(transcript)} words")
+
+    # Align markers
+    print("  - Aligning markers...")
+    alignments = align_markers(manuscript_text, transcript, offset_seconds=offset)
+
+    # Report results
+    print()
+    print("Alignment results:")
+    unmatched = 0
+    for a in alignments:
+        if a.timestamp >= 0:
+            print(f"  [{a.marker_id}] @ {a.timestamp:.2f}s - \"{a.matched_phrase}...\"")
+        else:
+            print(f"  [{a.marker_id}] NOT FOUND - \"{a.matched_phrase}...\"")
+            unmatched += 1
+
+    if unmatched > 0:
+        print(f"\nWarning: {unmatched} markers could not be aligned")
+
+    # Save aligned transcript.csv
+    output_path = project_path / "transcript.csv"
+    save_aligned_transcript(alignments, transcript, output_path)
+    print(f"\nSaved: {output_path}")
+
+    return 0
+
+
 if __name__ == "__main__":
     sys.exit(main())