Demonstrates the full pipeline: load audio → preprocess (prompt & target) → SVS inference with melody control → save output. Co-authored-by: Cursor <cursoragent@cursor.com>
