fix(*): decrease the row group size for audio datasets

By default, the row group size is set to 100 rows, which is too large
for the Hugging Face Dataset Viewer. This change reduces that number
to 20 rows. See this issue for more details:

https://github.com/google-deepmind/librispeech-long/issues/1

README.md

@@ -17,19 +17,19 @@ dataset_info:
     dtype: float64
   splits:
   - name: dev_clean
-    num_bytes: 1809240192.0
+    num_bytes: 1809224578.0
     num_examples: 295
   - name: dev_other
-    num_bytes: 1070584721.0
+    num_bytes: 1070577217.0
     num_examples: 188
   - name: test_clean
-    num_bytes: 1677792644.0
+    num_bytes: 1677717130.0
     num_examples: 270
   - name: test_other
-    num_bytes: 1195252663.0
+    num_bytes: 1195726954.0
     num_examples: 207
-  download_size: 5752870220
-  dataset_size: 5752870220
+  download_size: 5753245879
+  dataset_size: 5753245879
 configs:
 - config_name: librispeech_long
   data_files:

librispeech_long/dev_clean-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bca0ad12dd47d7f64eea2f365d39180217a14bcf9515cca16ad5bbc5d865b900
-size 1809240192
+oid sha256:a41961898641a1fde675bd71f0e30accc17ec43c5d77120323f0554867c63681
+size 1809224578

librispeech_long/dev_other-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1228721e30b2d72a8016a2d7ebe50cd525f2fc87d5ee29c94ee6c34ecfc65efd
-size 1070584721
+oid sha256:ce5c1aa12e33ace577e7ca8456d4aafa37dd3b6ff8b9098e676cf59caa49fe36
+size 1070577217

librispeech_long/test_clean-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2852d1108dff48cf05401327ae7dd49f11cc2f86678b02617e078c647514e09b
-size 1677792644
+oid sha256:8f16e89e3a3b69d644ee4844b8a66df4edffd119fd565a2629d3466d6f3d0348
+size 1677717130

librispeech_long/test_other-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fe9ecfafcebfb8c06af6764ec393f470a228dde4af19ec4d8f8768cb7199732
-size 1195252663
+oid sha256:f11c0e2fc9945de274c81303bbd253c592f33680a60bfe875fc3509ce78b2fd1
+size 1195726954

process.py

@@ -28,11 +28,14 @@ from dataclasses import dataclass
 from pathlib import Path
 from typing import Dict, List, Optional, Tuple
 
+import datasets
 from datasets import Audio, Dataset, Features, Value
 from tqdm import tqdm
 
 # --- Configuration ---
 
+datasets.config.PARQUET_ROW_GROUP_SIZE_FOR_AUDIO_DATASETS = 20
+
 # Config name on the Hub.
 CONFIG_NAME = "librispeech_long"
 
@@ -217,6 +220,7 @@ def build_parquet_dataset(rows: List[Row]) -> Dataset:
         }
         for r in rows
     ]
+
     return Dataset.from_list(data_list, features=features)