fix

@@ -27,6 +27,13 @@ python scripts/aliyun_dna/evaluate_aliyun_dna.py \
     --queries composition_testset_20/queries.csv \
     --out results/aliyun_dna_eval.csv \
     --duplicate-threshold 0.8
+
+# 测试翻唱库
+python scripts/aliyun_dna/evaluate_aliyun_dna.py \
+      --queries downloads_db/queries.csv \
+      --out results/dna_eval.csv \
+      --db-id f08aee1806c14df782e03cc1ab93f1ea \
+      --concurrency 4
 """
 
 import argparse

@@ -54,6 +54,42 @@ logger = logging.getLogger(__name__)
 
 DB_DSN = "postgresql://postgres:postgres@localhost:5432/embed_db"
 
+# 正样本变体：对主版本做轻度变换，模拟不同编码/剪辑场景
+# (variant_name, ffmpeg_extra_args_or_None)
+POSITIVE_VARIANTS: list[tuple[str, list[str] | None]] = [
+    # 转码为 128kbps MP3，模拟不同平台压缩率
+    ("mp3_128k", ["-codec:a", "libmp3lame", "-b:a", "128k"]),
+    # 重采样至 22050Hz，模拟不同采样率上传
+    ("resample_22k", ["-ar", "22050", "-ac", "1"]),
+    # 去掉前后各 5%，模拟用户剪掉片头片尾
+    ("trim_5pct", None),
+]
+
+
+def _make_variant(src: Path, dst: Path, ffmpeg_args: list[str] | None, duration: float | None = None) -> bool:
+    """生成单个正样本变体，返回是否成功。"""
+    import subprocess
+    if ffmpeg_args is None:
+        # trim_5pct：需要先探测时长
+        if duration is None:
+            probe = subprocess.run(
+                ["ffprobe", "-v", "error", "-show_entries", "format=duration",
+                 "-of", "default=noprint_wrappers=1:nokey=1", str(src)],
+                capture_output=True, text=True,
+            )
+            try:
+                duration = float(probe.stdout.strip())
+            except ValueError:
+                return False
+        ss = duration * 0.05
+        t = duration * 0.90
+        cmd = ["ffmpeg", "-y", "-i", str(src),
+               "-ss", f"{ss:.3f}", "-t", f"{t:.3f}", "-ac", "1", str(dst)]
+    else:
+        cmd = ["ffmpeg", "-y", "-i", str(src), *ffmpeg_args, "-ac", "1", str(dst)]
+    result = subprocess.run(cmd, capture_output=True)
+    return result.returncode == 0
+
 # fetch_records 查询返回的列，也是 --export-csv / --from-csv 的 CSV 格式
 RECORDS_FIELDS = [
     "record_id", "song_id", "song_name", "original_singer",
@@ -284,6 +320,8 @@ def main() -> None:
     ap.add_argument("--timeout", type=int, default=60, help="单文件下载超时秒数（默认 60）")
     ap.add_argument("--overwrite", action="store_true", help="覆盖已存在的文件")
     ap.add_argument("--dsn", default=DB_DSN, help="PostgreSQL 连接串")
+    ap.add_argument("--no-positive-variants", action="store_true",
+                    help="不生成正样本变体（mp3_128k/resample_22k/trim_5pct），只用 self 查询")
     ap.add_argument(
         "--export-csv", metavar="FILE",
         help="只导出查询结果到 CSV 后退出，不执行下载（在本地机器上运行）",
@@ -440,8 +478,11 @@ def main() -> None:
     ok_results = [r for r in results if r.get("status") == "ok" and r.get("local_path")]
 
     ref_rows, query_rows = [], []
+    variants_dir = output_dir / "variants"
+
     for r in ok_results:
         if int(r["is_main_version"]) == 1:
+            src = Path(r["local_path"])
             ref_rows.append({
                 "song_id": r["song_id"],
                 "audio_path": r["local_path"],
@@ -453,21 +494,39 @@ def main() -> None:
                 "genre": r["genre"],
                 "language_tag": r["language_tag"],
             })
-            # 正样本：主版本自身查询（验证入库是否成功）
-            query_rows.append({
-                "song_id": r["song_id"],
-                "audio_path": r["local_path"],
-                "variant": "self",
-                "sample_class": "positive",
-                "expected_song_id": r["song_id"],
-                "expected": "duplicate",
-                "song_name": r["song_name"],
-                "original_singer": r["original_singer"],
-                "singer_name": r["original_singer"],
-                "platform_name": r["platform_name"],
-                "duration": r["duration"],
-                "is_main_version": r["is_main_version"],
-            })
+
+            def _query_row(path, variant):
+                return {
+                    "song_id": r["song_id"],
+                    "audio_path": str(path),
+                    "variant": variant,
+                    "sample_class": "positive",
+                    "expected_song_id": r["song_id"],
+                    "expected": "duplicate",
+                    "song_name": r["song_name"],
+                    "original_singer": r["original_singer"],
+                    "singer_name": r["original_singer"],
+                    "platform_name": r["platform_name"],
+                    "duration": r["duration"],
+                    "is_main_version": r["is_main_version"],
+                }
+
+            # self：原始文件直接作为查询（验证基本入库）
+            query_rows.append(_query_row(src, "self"))
+
+            # 轻度变体正样本
+            if not args.no_positive_variants:
+                variants_dir.mkdir(parents=True, exist_ok=True)
+                duration = float(r["duration"]) if r.get("duration") else None
+                for vname, ffargs in POSITIVE_VARIANTS:
+                    ext = ".mp3" if vname == "mp3_128k" else src.suffix
+                    dst = variants_dir / f"{r['song_id']}_{vname}{ext}"
+                    if not dst.exists():
+                        ok = _make_variant(src, dst, ffargs, duration)
+                        if not ok:
+                            logger.warning("变体生成失败，跳过: song_id=%s %s", r["song_id"], vname)
+                            continue
+                    query_rows.append(_query_row(dst, vname))
         else:
             # 负样本：翻唱/其他版本，不应匹配到 DNA 库中的主版本
             platform = r["platform_name"] or "unknown"