Refactor librispeech unittest (#1140)

test/torchaudio_unittest/datasets/librispeech_test.py

 import os
 from pathlib import Path
 
-from torchaudio.datasets import librispeech
-
 from torchaudio_unittest.common_utils import (
     TempDirMixin,
     TorchaudioTestCase,
@@ -11,8 +9,10 @@ from torchaudio_unittest.common_utils import (
     normalize_wav,
 )
 
+from torchaudio.datasets import librispeech
+
 # Used to generate a unique utterance for each dummy audio file
-NUMBERS = [
+_NUMBERS = [
     'ZERO',
     'ONE',
     'TWO',
@@ -26,17 +26,13 @@ NUMBERS = [
 ]
 
 
-class TestLibriSpeech(TempDirMixin, TorchaudioTestCase):
-    backend = 'default'
-
-    root_dir = None
-    samples = []
-
-    @classmethod
-    def setUpClass(cls):
-        cls.root_dir = cls.get_base_temp_dir()
+def get_mock_dataset(root_dir):
+    """
+    root_dir: directory to the mocked dataset
+    """
+    mocked_data = []
     dataset_dir = os.path.join(
-            cls.root_dir, librispeech.FOLDER_IN_ARCHIVE, librispeech.URL
+        root_dir, librispeech.FOLDER_IN_ARCHIVE, librispeech.URL
     )
     os.makedirs(dataset_dir, exist_ok=True)
     sample_rate = 16000  # 16kHz
@@ -56,7 +52,7 @@ class TestLibriSpeech(TempDirMixin, TorchaudioTestCase):
                 path = os.path.join(chapter_path, filename)
 
                 utterance = ' '.join(
-                        [NUMBERS[x] for x in [speaker_id, chapter_id, utterance_id]]
+                    [_NUMBERS[x] for x in [speaker_id, chapter_id, utterance_id]]
                 )
                 trans_content.append(
                     f'{speaker_id}-{chapter_id}-{utterance_id:04d} {utterance}'
@@ -78,7 +74,7 @@ class TestLibriSpeech(TempDirMixin, TorchaudioTestCase):
                     chapter_id,
                     utterance_id
                 )
-                    cls.samples.append(sample)
+                mocked_data.append(sample)
 
                 seed += 1
 
@@ -86,6 +82,19 @@ class TestLibriSpeech(TempDirMixin, TorchaudioTestCase):
             trans_path = os.path.join(chapter_path, trans_filename)
             with open(trans_path, 'w') as f:
                 f.write('\n'.join(trans_content))
+    return mocked_data
+
+
+class TestLibriSpeech(TempDirMixin, TorchaudioTestCase):
+    backend = 'default'
+
+    root_dir = None
+    samples = []
+
+    @classmethod
+    def setUpClass(cls):
+        cls.root_dir = cls.get_base_temp_dir()
+        cls.samples = get_mock_dataset(cls.root_dir)
 
     @classmethod
     def tearDownClass(cls):