Update BirdSet.py

BirdSet.py

@@ -1,4 +1,4 @@
-# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+# Copyright 2025 The HuggingFace Datasets Authors and the current dataset script contributor.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -11,13 +11,14 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""BirdSet: The General Avian Monitoring Evaluation Benchmark"""
+"""BirdSet: A Large-Scale Dataset for Audio Classification in Avian Bioacoustics"""
 
 import os
 import datasets
 import pandas as pd
 from tqdm.auto import tqdm
 import tarfile
+from pathlib import Path
 
 from . import classes
 
@@ -34,18 +35,28 @@ from .descriptions import _NIPS4BPLUS_CITATION, _NIPS4BPLUS_DESCRIPTION, \
 
 #############################################
 _BIRDSET_CITATION = """\
-    @article{birdset,
-        title = {BirdSet: A Multi-Task Benchmark For Avian Diversity Monitoring},
-        author={anonymous},
-        year={2024}
+    @misc{rauch2025birdsetlargescaledatasetaudio,
+          title={BirdSet: A Large-Scale Dataset for Audio Classification in Avian Bioacoustics}, 
+          author={Lukas Rauch and Raphael Schwinger and Moritz Wirth and René Heinrich and Denis Huseljic and Marek Herde and Jonas Lange and Stefan Kahl and Bernhard Sick and Sven Tomforde and Christoph Scholz},
+          year={2025},
+          eprint={2403.10380},
+          archivePrefix={arXiv},
+          primaryClass={cs.SD},
+          url={https://arxiv.org/abs/2403.10380}, 
     }
 """
 _BIRDSET_DESCRIPTION = """\
-    This dataset offers a unified, well-structured platform for avian bioacoustics and consists of various tasks. \
-    By creating a set of tasks, BirdSet enables an overall performance score for models and uncovers their limitations \
-    in certain areas.
-    Note that each BirdSet dataset has its own citation. Please see the source to get the correct citation for each 
-    contained dataset. 
+    Deep learning (DL) has greatly advanced audio classification, 
+    yet the field is limited by the scarcity of large-scale benchmark datasets that have propelled progress in other domains. 
+    While AudioSet is a pivotal step to bridge this gap as a universal-domain dataset, its restricted accessibility and 
+    limited range of evaluation use cases challenge its role as the sole resource. Therefore, we introduce BirdSet, 
+    a large-scale benchmark dataset for audio classification focusing on avian bioacoustics. 
+    BirdSet surpasses AudioSet with over 6,800 recording hours (+17%) from nearly 10,000 classes (x18) for training and more 
+    than 400 hours (x7) across eight strongly labeled evaluation datasets. It serves as a versatile resource for use 
+    cases such as multi-label classification, covariate shift or self-supervised learning. We benchmark six well-known 
+    DL models in multi-label classification across three distinct training scenarios and outline further evaluation use 
+    cases in audio classification. We host our dataset on Hugging Face for easy accessibility and offer an extensive 
+    codebase to reproduce our results. 
 """
 
 base_url = "https://huggingface.co/datasets/DBD-research-group/BirdSet/resolve/data"
@@ -67,7 +78,7 @@ def _extract_all_to_same_folder(tar_path, output_dir):
     return output_dir
 
 
-def _extract_and_delete(dl_dir: dict) -> dict:
+def _extract_and_delete(dl_dir: dict, cache_dir: str = None) -> dict:
     """extracts downloaded files and deletes the archive file immediately, with progress bar.
     only the processed archive and its content are saved at the same time."""
     audio_paths = {name: [] for name, data in dl_dir.items() if isinstance(data, list)}
@@ -77,8 +88,8 @@ def _extract_and_delete(dl_dir: dict) -> dict:
 
         # extract and immediately delete archives
         for path in tqdm(data, f"Extracting {name} split"):
-            head, tail = os.path.split(path)
-            output_dir = os.path.join(head, "extracted", tail)
+            directory, filename = os.path.split(path)
+            output_dir = os.path.join(cache_dir or directory, "extracted", filename.split(".")[0])
             #audio_path = dl_manager.extract(path) # if all archive files are without subfolders this works just fine
             audio_path = _extract_all_to_same_folder(path, output_dir)
             os.remove(path)
@@ -105,7 +116,7 @@ class BirdSetConfig(datasets.BuilderConfig):
         features = datasets.Features({
             "audio": datasets.Audio(sampling_rate=32_000, mono=True, decode=False),
             "filepath": datasets.Value("string"),
-            "start_time": datasets.Value("float64"),  # can be changed to timestamp later
+            "start_time": datasets.Value("float64"),
             "end_time": datasets.Value("float64"),
             "low_freq": datasets.Value("int64"),
             "high_freq": datasets.Value("int64"),
@@ -482,7 +493,7 @@ class BirdSet(datasets.GeneratorBasedBuilder):
             })
 
         # custom extraction that deletes archives right after extraction
-        audio_paths = _extract_and_delete(dl_dir) if not dl_manager.is_streaming else None
+        audio_paths = _extract_and_delete(dl_dir, dl_manager.download_config.cache_dir) if not dl_manager.is_streaming else None
 
         # construct split generators
         # assumes every key in dl_dir of NAME also has meta_NAME