Update dataset loading script

jeli-asr.py

@@ -15,7 +15,6 @@ limitations under the License.
 """
 # jeli_asr.py
 import csv
-import os
 
 import datasets
 from datasets import Split, SplitGenerator
@@ -123,10 +122,12 @@ class JeliASR(datasets.GeneratorBasedBuilder):
         Read metadata.csv row-by-row, filter by subset (based on config) & split,
         then download each audio file from the Hub and yield its local path.
         """
+        audios_to_download = []
+        metadata_dict = {}
+
         with open(metadata_path, "r", encoding="utf-8") as f:
             reader = csv.DictReader(f)
-            idx = 0
-            for row in reader:
+            for idx, row in enumerate(reader):
                 file_path = row["file_name"]  # e.g. "bam-asr-oza/train/.../audio.wav"
 
                 # 1) Filter by config
@@ -144,15 +145,23 @@ class JeliASR(datasets.GeneratorBasedBuilder):
 
                 # Build the raw URL for the audio
                 audio_url = f"{_BASE_URL}/{file_path}"
-                # => e.g. "https://huggingface.co/datasets/RobotsMali/jeli-asr/resolve/main/bam-asr-oza/train/split1/oza75-bam-asr-0.wav"
-
-                # Download it. dl_manager returns the local path in the cache.
-                local_audio_path = dl_manager.download(audio_url)
+                audios_to_download.append(audio_url)
 
-                yield idx, {
-                    "audio": local_audio_path,  # local path for datasets.Audio
+                # Store metadata in dictionary
+                metadata_dict[audio_url] = {
                     "duration": float(row["duration"]),
                     "bam": row["bam"],
                     "french": row["french"],
                 }
-                idx += 1
+
+        # Download the audios. dl_manager returns the local paths in the cache.
+        local_audio_paths = dl_manager.download(audios_to_download)
+        for idx, audio_url in enumerate(audios_to_download):
+            local_audio_path = local_audio_paths[idx]
+            # Yield the example
+            yield idx, {
+                "audio": local_audio_path,  # local path for datasets.Audio
+                "duration": metadata_dict[audio_url]["duration"],
+                "bam": metadata_dict[audio_url]["bam"],
+                "french": metadata_dict[audio_url]["french"],
+            }