Add audio support for croissant builder.

PiperOrigin-RevId: 783663338

Author: The TensorFlow Datasets Authors

tensorflow_datasets/core/dataset_builders/croissant_builder.py

@@ -48,6 +48,7 @@
 from tensorflow_datasets.core import download
 from tensorflow_datasets.core import split_builder as split_builder_lib
 from tensorflow_datasets.core import splits as splits_lib
+from tensorflow_datasets.core.features import audio_feature
 from tensorflow_datasets.core.features import bounding_boxes
 from tensorflow_datasets.core.features import feature as feature_lib
 from tensorflow_datasets.core.features import features_dict
@@ -177,6 +178,10 @@ def datatype_converter(
     feature = bounding_boxes.BBoxFeature(
         doc=field.description, bbox_format=None
     )
+  elif field_data_type == mlc.DataType.AUDIO_OBJECT:
+    feature = audio_feature.Audio(
+        doc=field.description, sample_rate=field.source.sampling_rate
+    )
   else:
     raise ValueError(f'Unknown data type: {field_data_type}.')
 

tensorflow_datasets/core/dataset_builders/croissant_builder_test.py

@@ -20,6 +20,7 @@
 from tensorflow_datasets import testing
 from tensorflow_datasets.core import file_adapters
 from tensorflow_datasets.core.dataset_builders import croissant_builder
+from tensorflow_datasets.core.features import audio_feature
 from tensorflow_datasets.core.features import bounding_boxes
 from tensorflow_datasets.core.features import features_dict
 from tensorflow_datasets.core.features import image_feature
@@ -150,6 +151,14 @@ def test_simple_datatype_converter(
             image_feature.Image,
             None,
         ),
+        (
+            mlc.Field(
+                data_types=mlc.DataType.AUDIO_OBJECT,
+                description="Audio feature",
+            ),
+            audio_feature.Audio,
+            None,
+        ),
         (
             mlc.Field(
                 data_types=mlc.DataType.BOUNDING_BOX,