handle various input types e.g across different gradio versions

offline_pipeline.py

@@ -19,6 +19,26 @@ from utils import (
 SDK_OFFLINE = SDKWrapper()
 
 
+def _extract_uploaded_path(sample_input: Any) -> str | None:
+    if sample_input is None:
+        return None
+    if isinstance(sample_input, str):
+        return sample_input
+
+    for attr in ("path", "name"):
+        value = getattr(sample_input, attr, None)
+        if isinstance(value, str):
+            return value
+
+    if isinstance(sample_input, dict):
+        for key in ("path", "name"):
+            value = sample_input.get(key)
+            if isinstance(value, str):
+                return value
+
+    return None
+
+
 def _safe_progress(progress: gr.Progress, value: float, desc: str) -> None:
     progress(max(0.0, min(1.0, value)), desc=desc)
 
@@ -236,15 +256,16 @@ def run_offline_pipeline(
 
 
 def load_local_file(
-    sample_path: str,
+    sample_path: Any,
     normalize: bool = True,
 ) -> tuple[np.ndarray | None, str, tuple | None, int | None]:
+    sample_path = _extract_uploaded_path(sample_path)
     if not sample_path or not os.path.exists(sample_path):
         return None, "", None, None
 
     if os.path.getsize(sample_path) > 5 * 1024 * 1024:
         gr.Warning("File size exceeds 5 MB limit. Please upload a smaller file.")
-        raise ValueError("Uploaded file exceeds the 5 MB size limit.")
+        return None, "", None, None
 
     new_sample_stem = os.path.splitext(os.path.basename(sample_path))[0]
     y, sample_rate = librosa.load(sample_path, sr=None, mono=True)
@@ -274,4 +295,4 @@ def load_file_from_dataset(
     if y.ndim > 1:
         y = np.mean(y, axis=0)
     gradio_audio = to_gradio_audio(y, sample_rate)
-    return gradio_audio, y, new_sample_stem, sample_rate
+    return gradio_audio, y, new_sample_stem, sample_rate