Update scripts/render_results.py

scripts/render_results.py

@@ -10,7 +10,6 @@ when the README is rendered as the model card.
 from __future__ import annotations
 
 import json
-from itertools import zip_longest
 from pathlib import Path
 
 ROOT = Path(__file__).resolve().parent.parent
@@ -50,17 +49,39 @@ DATASETS = [
         "link":  "https://huggingface.co/datasets/laion/freesound-commercially-permissive-subset-with-captions",
         "blurb": "Curated commercially-permissive Freesound clips — 10 "
                  "examples that the AST router classified as something "
-                 "other than speech or music. Each clip is shown alongside "
-                 "the dataset's own upstream caption for comparison.",
-        "with_upstream": True,
+                 "other than speech or music, so they are routed through "
+                 "the general-purpose sound-effect captioner.",
     },
 ]
 
 
+def _escape_md_emphasis(text: str) -> str:
+    """Escape characters that would otherwise toggle Markdown emphasis.
+
+    Hugging Face renders blockquotes (`> ...`) in italic by default, so
+    a stray ``*`` or ``_`` inside a caption flips the italic off mid-text
+    and the rest of the paragraph appears in upright type. Backslash-
+    escaping these (and ``~`` for strikethrough) keeps the blockquote
+    italic from end to end.
+    """
+    return (
+        text.replace("\\", "\\\\")
+            .replace("*", "\\*")
+            .replace("_", "\\_")
+            .replace("~", "\\~")
+    )
+
+
+def _quote_caption(text: str) -> str:
+    """Render multi-line caption text as a single blockquote."""
+    text = _escape_md_emphasis(text.strip())
+    lines = text.split("\n")
+    return "\n".join(f"> {ln}" if ln.strip() else ">" for ln in lines)
+
+
 def render_clip(item: dict) -> str:
     payload  = item["payload"]
     audio    = item["audio_url"]
-    upstream = item.get("upstream")
     ds       = item["dataset"]
 
     out: list[str] = []
@@ -75,21 +96,6 @@ def render_clip(item: dict) -> str:
     )
     out.append("")
 
-    if upstream is not None:
-        cap = upstream.get("comprehensive_caption")
-        fid = upstream.get("freesound_id")
-        if cap:
-            out.append("**Upstream dataset caption:**")
-            out.append("")
-            if fid:
-                out.append(
-                    f"> {cap}  \n> _(Freesound id "
-                    f"[`{fid}`](https://freesound.org/s/{fid}/))_"
-                )
-            else:
-                out.append(f"> {cap}")
-            out.append("")
-
     out.append("**AudioSet top-3 predictions** (MIT AST):")
     out.append("")
     out.append("| # | Label | Confidence |")
@@ -106,24 +112,24 @@ def render_clip(item: dict) -> str:
         if "voice_tags" in ann:
             out.append("**`laion/voice-tagging-whisper` — voice tags:**")
             out.append("")
-            out.append(f"> {ann['voice_tags']}")
+            out.append(_quote_caption(ann["voice_tags"]))
             out.append("")
         if "bud_e_speech_caption" in ann:
             out.append("**`laion/BUD-E-Whisper_V1.2` — speech caption:**")
             out.append("")
-            out.append(f"> {ann['bud_e_speech_caption']}")
+            out.append(_quote_caption(ann["bud_e_speech_caption"]))
             out.append("")
     elif payload["route"] == "music":
         if "music_caption" in ann:
             out.append("**`laion/music-whisper` — music caption:**")
             out.append("")
-            out.append(f"> {ann['music_caption']}")
+            out.append(_quote_caption(ann["music_caption"]))
             out.append("")
     else:  # sfx
         if "sound_effect_caption" in ann:
             out.append("**`laion/sound-effect-captioning-whisper` — sound caption:**")
             out.append("")
-            out.append(f"> {ann['sound_effect_caption']}")
+            out.append(_quote_caption(ann["sound_effect_caption"]))
             out.append("")
 
     out.append("---")
@@ -135,12 +141,6 @@ def collect_clips() -> list[dict]:
     """Load every sample JSON sidecar and return a flat list of items."""
     items: list[dict] = []
     for ds in DATASETS:
-        upstream_map: dict[str, dict] = {}
-        if ds.get("with_upstream"):
-            up_path = SAMPLES / ds["short"] / "upstream_captions.json"
-            if up_path.exists():
-                upstream_map = json.loads(up_path.read_text())
-
         json_files = [
             jf for jf in sorted((SAMPLES / ds["short"]).glob("*.json"))
             if jf.name != "upstream_captions.json"
@@ -153,7 +153,6 @@ def collect_clips() -> list[dict]:
                 "dataset":  ds,
                 "payload":  payload,
                 "audio_url": audio_url,
-                "upstream": upstream_map.get(audio_basename),
             })
     return items