Add Weldy's calltype dataset and Anuraset.

PiperOrigin-RevId: 553610183

chirp/data/soundscapes/dataset_fns.py

@@ -209,3 +209,53 @@ def load_powdermill_annotations(annotations_path: epath.Path) -> pd.DataFrame:
   )
   segments = annotations.annotations_to_dataframe(annos)
   return segments
+
+
+def load_weldy_annotations(annotations_path: epath.Path) -> pd.DataFrame:
+  """Loads a dataframe of all annotations from the Weldy Calltype dataset."""
+  filename_fn = lambda _, row: 'annotated_recordings/' + row['file'].strip()
+  start_time_fn = lambda row: float(row['start'])
+  end_time_fn = lambda row: float(row['end'])
+  filter_fn = lambda row: False
+  class_fn = lambda row: (  # pylint: disable=g-long-lambda
+      row['label']
+      .replace('unk', 'unknown')
+      .replace('impossible', 'unknown')
+      .replace('unknown_chip', 'unknown')
+      .split(' ')
+  )
+  annos = annotations.read_dataset_annotations_csvs(
+      [epath.Path(annotations_path)],
+      filename_fn=filename_fn,
+      namespace='weldy_calltype',
+      class_fn=class_fn,
+      start_time_fn=start_time_fn,
+      end_time_fn=end_time_fn,
+      filter_fn=filter_fn,
+  )
+  segments = annotations.annotations_to_dataframe(annos)
+  return segments
+
+
+def load_anuraset_annotations(annotations_path: epath.Path) -> pd.DataFrame:
+  """Loads a dataframe of all annotations."""
+  filename_fn = lambda _, row: os.path.join(  # pylint: disable=g-long-lambda
+      row['filename'].split('_')[0], row['filename'].strip()
+  )
+  start_time_fn = lambda row: float(row['start_time_s'])
+  end_time_fn = lambda row: float(row['end_time_s'])
+  # There are a few SPECIES_LALSE labels which according to the authors should
+  # be ignored.
+  filter_fn = lambda row: '_LALSE' in row['label']
+  class_fn = lambda row: row['label'].split(' ')
+  annos = annotations.read_dataset_annotations_csvs(
+      [epath.Path(annotations_path)],
+      filename_fn=filename_fn,
+      namespace='anuraset',
+      class_fn=class_fn,
+      start_time_fn=start_time_fn,
+      end_time_fn=end_time_fn,
+      filter_fn=filter_fn,
+  )
+  segments = annotations.annotations_to_dataframe(annos)
+  return segments

chirp/data/soundscapes/soundscapes.py

@@ -335,6 +335,30 @@ class Soundscapes(bird_taxonomy.BirdTaxonomy):
           ),
           class_list_name='peru',
       ),
+      SoundscapesConfig(
+          name='weldy_calltype_full_length',
+          audio_glob='weldy_calltype/annotated_recordings/*.wav',
+          annotation_load_fn=dataset_fns.load_weldy_annotations,
+          annotation_filename='annotations.csv',
+          keep_unknown_annotation=True,
+          description=(
+              'Full-length annotated bird call types from the PNW. '
+              'https://zenodo.org/record/8047850'
+          ),
+          class_list_name='weldy_calltype',
+      ),
+      SoundscapesConfig(
+          name='anuraset_full_length',
+          audio_glob='anuraset/raw_data/*/*.wav',
+          annotation_load_fn=dataset_fns.load_anuraset_annotations,
+          annotation_filename='annotations.csv',
+          keep_unknown_annotation=True,
+          description=(
+              'Full-length annotated frog vocalizations. '
+              'https://zenodo.org/record/8056090'
+          ),
+          class_list_name='anuraset',
+      ),
   ]
 
   def _info(self) -> tfds.core.DatasetInfo:

chirp/taxonomy/namespace_db.py

@@ -48,24 +48,33 @@ def validate_taxonomy_database(taxonomy_database: TaxonomyDatabase) -> None:
   """
   namespaces = taxonomy_database.namespaces
 
-  for _, mapping in taxonomy_database.mappings.items():
+  for mapping_name, mapping in taxonomy_database.mappings.items():
     if (
         set(mapping.mapped_pairs.keys())
         - namespaces[mapping.source_namespace].classes
     ):
-      raise ValueError("unknown class in source")
+      raise ValueError(
+          f"Mapping {mapping_name} contains a source class not in "
+          f"the namespace ({mapping.source_namespace})."
+      )
     if (
         set(mapping.mapped_pairs.values())
         - namespaces[mapping.target_namespace].classes
     ):
-      raise ValueError("unknown class in target")
+      raise ValueError(
+          f"Mapping {mapping_name} contains a target class not in "
+          f"the namespace ({mapping.source_namespace})."
+      )
 
-  for _, class_list in taxonomy_database.class_lists.items():
+  for class_name, class_list in taxonomy_database.class_lists.items():
     classes = class_list.classes
     if set(classes) - namespaces[class_list.namespace].classes > {
         namespace.UNKNOWN_LABEL
     }:
-      raise ValueError("unknown class in class list")
+      raise ValueError(
+          f"ClassList {class_name} contains a class not in "
+          f"the namespace ({class_list.namespace})."
+      )
 
 
 def load_taxonomy_database(