From 2c71d4c1847ed22e273470a646470424af16552a Mon Sep 17 00:00:00 2001
From: Ethanfel <ethan.fel@ts-pc.fr>
Date: Thu, 9 Apr 2026 14:09:43 +0200
Subject: [PATCH] feat: add SelvaDatasetLoader node

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 nodes/selva_dataset_pipeline.py | 42 +++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)

diff --git a/nodes/selva_dataset_pipeline.py b/nodes/selva_dataset_pipeline.py
index d25f4e1..d1d78dd 100644
--- a/nodes/selva_dataset_pipeline.py
+++ b/nodes/selva_dataset_pipeline.py
@@ -24,3 +24,45 @@ from .utils import SELVA_CATEGORY
 AUDIO_DATASET = "AUDIO_DATASET"
 
 _AUDIO_EXTS = {".wav", ".flac", ".mp3", ".ogg", ".aac", ".m4a"}
+
+
+class SelvaDatasetLoader:
+    """Load all audio files in a folder into an in-memory AUDIO_DATASET."""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "folder": ("STRING", {
+                    "default": "",
+                    "tooltip": "Absolute path to folder containing audio files. Searched recursively.",
+                }),
+            }
+        }
+
+    RETURN_TYPES  = (AUDIO_DATASET,)
+    RETURN_NAMES  = ("dataset",)
+    FUNCTION      = "load"
+    CATEGORY      = SELVA_CATEGORY
+    DESCRIPTION   = "Load all audio files from a folder into memory as an AUDIO_DATASET."
+
+    def load(self, folder: str):
+        folder = Path(folder.strip())
+        if not folder.exists():
+            raise FileNotFoundError(f"[DatasetLoader] Folder not found: {folder}")
+
+        files = [f for f in folder.rglob("*") if f.suffix.lower() in _AUDIO_EXTS]
+        if not files:
+            raise RuntimeError(f"[DatasetLoader] No audio files found in {folder}")
+
+        dataset = []
+        for f in sorted(files):
+            try:
+                wav, sr = torchaudio.load(str(f))          # [C, L]
+                wav = wav.unsqueeze(0).float()              # [1, C, L]
+                dataset.append({"waveform": wav, "sample_rate": sr, "name": f.stem})
+            except Exception as e:
+                print(f"[DatasetLoader] Skipping {f.name}: {e}", flush=True)
+
+        print(f"[DatasetLoader] Loaded {len(dataset)} clips from {folder}", flush=True)
+        return (dataset,)