make style and fix import (remove torchaudio dependency import)

src/transformers/__init__.py

@@ -146,6 +146,7 @@
     "models.audio_spectrogram_transformer": [
         "AUDIO_SPECTROGRAM_TRANSFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "ASTConfig",
+        "ASTFeatureExtractor",
     ],
     "models.auto": [
         "ALL_PRETRAINED_CONFIG_ARCHIVE_MAP",
@@ -514,6 +515,7 @@
     "models.speech_to_text": [
         "SPEECH_TO_TEXT_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "Speech2TextConfig",
+        "Speech2TextFeatureExtractor",
         "Speech2TextProcessor",
     ],
     "models.speech_to_text_2": [
@@ -887,20 +889,6 @@
 else:
     _import_structure["convert_slow_tokenizer"] = ["SLOW_TO_FAST_CONVERTERS", "convert_slow_tokenizer"]
 
-# Speech-specific objects
-try:
-    if not is_speech_available():
-        raise OptionalDependencyNotAvailable()
-except OptionalDependencyNotAvailable:
-    from .utils import dummy_speech_objects
-
-    _import_structure["utils.dummy_speech_objects"] = [
-        name for name in dir(dummy_speech_objects) if not name.startswith("_")
-    ]
-else:
-    _import_structure["models.audio_spectrogram_transformer"].append("ASTFeatureExtractor")
-    _import_structure["models.speech_to_text"].append("Speech2TextFeatureExtractor")
-
 # Tensorflow-text-specific objects
 try:
     if not is_tensorflow_text_available():
@@ -4275,6 +4263,7 @@
     from .models.audio_spectrogram_transformer import (
         AUDIO_SPECTROGRAM_TRANSFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP,
         ASTConfig,
+        ASTFeatureExtractor,
     )
     from .models.auto import (
         ALL_PRETRAINED_CONFIG_ARCHIVE_MAP,
@@ -4624,6 +4613,7 @@
     from .models.speech_to_text import (
         SPEECH_TO_TEXT_PRETRAINED_CONFIG_ARCHIVE_MAP,
         Speech2TextConfig,
+        Speech2TextFeatureExtractor,
         Speech2TextProcessor,
     )
     from .models.speech_to_text_2 import (
@@ -4964,15 +4954,6 @@
     else:
         from .convert_slow_tokenizer import SLOW_TO_FAST_CONVERTERS, convert_slow_tokenizer
 
-    try:
-        if not is_speech_available():
-            raise OptionalDependencyNotAvailable()
-    except OptionalDependencyNotAvailable:
-        from .utils.dummy_speech_objects import *
-    else:
-        from .models.audio_spectrogram_transformer import ASTFeatureExtractor
-        from .models.speech_to_text import Speech2TextFeatureExtractor
-
     try:
         if not is_tensorflow_text_available():
             raise OptionalDependencyNotAvailable()

src/transformers/models/audio_spectrogram_transformer/__init__.py

@@ -13,14 +13,15 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_speech_available, is_torch_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available
 
 
 _import_structure = {
     "configuration_audio_spectrogram_transformer": [
         "AUDIO_SPECTROGRAM_TRANSFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "ASTConfig",
-    ]
+    ],
+    "feature_extraction_audio_spectrogram_transformer": ["ASTFeatureExtractor"],
 }
 
 try:
@@ -36,19 +37,13 @@
         "ASTPreTrainedModel",
     ]
 
-try:
-    if not is_speech_available():
-        raise OptionalDependencyNotAvailable()
-except OptionalDependencyNotAvailable:
-    pass
-else:
-    _import_structure["feature_extraction_audio_spectrogram_transformer"] = ["ASTFeatureExtractor"]
 
 if TYPE_CHECKING:
     from .configuration_audio_spectrogram_transformer import (
         AUDIO_SPECTROGRAM_TRANSFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP,
         ASTConfig,
     )
+    from .feature_extraction_audio_spectrogram_transformer import ASTFeatureExtractor
 
     try:
         if not is_torch_available():
@@ -63,14 +58,6 @@
             ASTPreTrainedModel,
         )
 
-    try:
-        if not is_speech_available():
-            raise OptionalDependencyNotAvailable()
-    except OptionalDependencyNotAvailable:
-        pass
-    else:
-        from .feature_extraction_audio_spectrogram_transformer import ASTFeatureExtractor
-
 
 else:
     import sys

src/transformers/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.py

@@ -26,6 +26,7 @@
 from ...feature_extraction_utils import BatchFeature
 from ...utils import TensorType, is_speech_available, logging
 
+
 if is_speech_available():
     import torchaudio.compliance.kaldi as ta_kaldi
 

src/transformers/models/speech_to_text/__init__.py

@@ -17,14 +17,14 @@
     OptionalDependencyNotAvailable,
     _LazyModule,
     is_sentencepiece_available,
-    is_speech_available,
     is_tf_available,
     is_torch_available,
 )
 
 
 _import_structure = {
     "configuration_speech_to_text": ["SPEECH_TO_TEXT_PRETRAINED_CONFIG_ARCHIVE_MAP", "Speech2TextConfig"],
+    "feature_extraction_speech_to_text": ["Speech2TextFeatureExtractor"],
     "processing_speech_to_text": ["Speech2TextProcessor"],
 }
 
@@ -36,14 +36,6 @@
 else:
     _import_structure["tokenization_speech_to_text"] = ["Speech2TextTokenizer"]
 
-try:
-    if not is_speech_available():
-        raise OptionalDependencyNotAvailable()
-except OptionalDependencyNotAvailable:
-    pass
-else:
-    _import_structure["feature_extraction_speech_to_text"] = ["Speech2TextFeatureExtractor"]
-
 try:
     if not is_tf_available():
         raise OptionalDependencyNotAvailable()
@@ -73,6 +65,7 @@
 
 if TYPE_CHECKING:
     from .configuration_speech_to_text import SPEECH_TO_TEXT_PRETRAINED_CONFIG_ARCHIVE_MAP, Speech2TextConfig
+    from .feature_extraction_speech_to_text import Speech2TextFeatureExtractor
     from .processing_speech_to_text import Speech2TextProcessor
 
     try:
@@ -83,14 +76,6 @@
     else:
         from .tokenization_speech_to_text import Speech2TextTokenizer
 
-    try:
-        if not is_speech_available():
-            raise OptionalDependencyNotAvailable()
-    except OptionalDependencyNotAvailable:
-        pass
-    else:
-        from .feature_extraction_speech_to_text import Speech2TextFeatureExtractor
-
     try:
         if not is_tf_available():
             raise OptionalDependencyNotAvailable()

src/transformers/models/speech_to_text/feature_extraction_speech_to_text.py

@@ -20,14 +20,16 @@
 
 import numpy as np
 import torch
-import torchaudio.compliance.kaldi as ta_kaldi
 
 from ...audio_utils import mel_filter_bank, spectrogram, window_function
 from ...feature_extraction_sequence_utils import SequenceFeatureExtractor
 from ...feature_extraction_utils import BatchFeature
 from ...utils import PaddingStrategy, TensorType, is_speech_available, logging
 
 
+if is_speech_available():
+    import torchaudio.compliance.kaldi as ta_kaldi
+
 logger = logging.get_logger(__name__)
 
 

tests/models/audio_spectrogram_transformer/test_feature_extraction_audio_spectrogram_transformer.py

@@ -134,7 +134,7 @@ def test_call(self):
         encoded_sequences_2 = feat_extract(np_speech_inputs, return_tensors="np").input_values
         for enc_seq_1, enc_seq_2 in zip(encoded_sequences_1, encoded_sequences_2):
             self.assertTrue(np.allclose(enc_seq_1, enc_seq_2, atol=1e-3))
-    
+
     @require_torch
     def test_double_precision_pad(self):
         import torch
@@ -174,7 +174,7 @@ def test_integration(self):
         input_values = feature_extractor(input_speech, return_tensors="pt").input_values
         self.assertEquals(input_values.shape, (1, 1024, 128))
         self.assertTrue(torch.allclose(input_values[0, 0, :30], EXPECTED_INPUT_VALUES, atol=1e-4))
-    
+
     def test_feat_extract_from_and_save_pretrained(self):
         feat_extract_first = self.feature_extraction_class(**self.feat_extract_dict)
 
@@ -201,16 +201,20 @@ def test_feat_extract_to_json_file(self):
 
 
 @require_torch
-@unittest.mock.patch("transformers.models.audio_spectrogram_transformer.feature_extraction_audio_spectrogram_transformer.is_speech_available", lambda: False)
+@unittest.mock.patch(
+    "transformers.models.audio_spectrogram_transformer.feature_extraction_audio_spectrogram_transformer.is_speech_available",
+    lambda: False,
+)
 class ASTFeatureExtractionWithoutTorchaudioTest(SequenceFeatureExtractionTestMixin, unittest.TestCase):
     feature_extraction_class = ASTFeatureExtractor
+
     def setUp(self):
         self.feat_extract_tester = ASTFeatureExtractionTester(self)
-        
+
     def test_using_audio_utils(self):
         # Tests that it uses audio_utils instead of torchaudio
         feat_extract = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict())
-        
+
         self.assertTrue(hasattr(feat_extract, "window"))
         self.assertTrue(hasattr(feat_extract, "mel_filters"))
 

tests/models/speech_to_text/test_feature_extraction_speech_to_text.py

@@ -22,13 +22,12 @@
 
 import numpy as np
 
+from transformers import Speech2TextFeatureExtractor
 from transformers.testing_utils import check_json_file_has_correct_format, require_torch, require_torchaudio
 
 from ...test_sequence_feature_extraction_common import SequenceFeatureExtractionTestMixin
 
 
-from transformers import Speech2TextFeatureExtractor
-
 global_rng = random.Random()
 
 
@@ -306,7 +305,9 @@ def test_feat_extract_to_json_file(self):
 
 
 @require_torch
-@unittest.mock.patch("transformers.models.speech_to_text.feature_extraction_speech_to_text.is_speech_available", lambda: False)
+@unittest.mock.patch(
+    "transformers.models.speech_to_text.feature_extraction_speech_to_text.is_speech_available", lambda: False
+)
 class Speech2TextFeatureExtractionWithoutTorchaudioTest(SequenceFeatureExtractionTestMixin, unittest.TestCase):
     feature_extraction_class = Speech2TextFeatureExtractor
 
@@ -320,7 +321,7 @@ def _check_zero_mean_unit_variance(self, input_vector):
     def test_using_audio_utils(self):
         # Tests that it uses audio_utils instead of torchaudio
         feat_extract = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict())
-        
+
         self.assertTrue(hasattr(feat_extract, "window"))
         self.assertTrue(hasattr(feat_extract, "mel_filters"))
 

tests/models/speech_to_text/test_processor_speech_to_text.py

@@ -18,13 +18,14 @@
 from pathlib import Path
 from shutil import copyfile
 
-from transformers import Speech2TextTokenizer, Speech2TextFeatureExtractor, Speech2TextProcessor
+from transformers import Speech2TextFeatureExtractor, Speech2TextProcessor, Speech2TextTokenizer
 from transformers.models.speech_to_text.tokenization_speech_to_text import VOCAB_FILES_NAMES, save_json
 from transformers.testing_utils import get_tests_dir, require_sentencepiece, require_torch, require_torchaudio
 from transformers.utils import FEATURE_EXTRACTOR_NAME
 
 from .test_feature_extraction_speech_to_text import floats_list
 
+
 SAMPLE_SP = get_tests_dir("fixtures/test_sentencepiece.model")