SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Feature Extractor: Wav2Vec2 & Speech2Text - Allow truncation + padding=longest (#13600)

Browse Source 
* correct

* add tests

* Update src/transformers/feature_extraction_sequence_utils.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
This commit is contained in:
Patrick von Platen
2021-09-16 20:02:54 +02:00
committed by GitHub
parent e59041684e
commit 4d5b4c7863
5 changed files with 91 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
tests/test_feature_extraction_speech_to_text.py
View File

@@ -189,10 +189,12 @@ class Speech2TextFeatureExtractionTest(SequenceFeatureExtractionTestMixin, unitt
self.assertTrue(np.all(np.abs(np.var(input_vector, axis=0) - 1) < var_tol))
_check_zero_mean_unit_variance(input_features[0][: fbank_feat_lengths[0]], var_tol)
self.assertTrue(input_features[0][fbank_feat_lengths[0] :].sum() < 1e-6)
_check_zero_mean_unit_variance(input_features[1][: fbank_feat_lengths[1]], var_tol)
self.assertTrue(input_features[0][fbank_feat_lengths[1] :].sum() < 1e-6)
_check_zero_mean_unit_variance(input_features[2][: fbank_feat_lengths[2]], var_tol)
def test_cepstral_mean_and_variance_normalization_trunc(self):
def test_cepstral_mean_and_variance_normalization_trunc_max_length(self):
feature_extractor = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict())
speech_inputs = [floats_list((1, x))[0] for x in range(800, 1400, 200)]
inputs = feature_extractor(
@@ -214,3 +216,49 @@ class Speech2TextFeatureExtractionTest(SequenceFeatureExtractionTestMixin, unitt
_check_zero_mean_unit_variance(input_features[0, : fbank_feat_lengths[0]])
_check_zero_mean_unit_variance(input_features[1])
_check_zero_mean_unit_variance(input_features[2])
def test_cepstral_mean_and_variance_normalization_trunc_longest(self):
feature_extractor = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict())
speech_inputs = [floats_list((1, x))[0] for x in range(800, 1400, 200)]
inputs = feature_extractor(
speech_inputs,
padding="longest",
max_length=4,
truncation=True,
return_tensors="np",
return_attention_mask=True,
)
input_features = inputs.input_features
attention_mask = inputs.attention_mask
fbank_feat_lengths = np.sum(attention_mask == 1, axis=1)
def _check_zero_mean_unit_variance(input_vector):
self.assertTrue(np.all(np.mean(input_vector, axis=0) < 1e-3))
self.assertTrue(np.all(np.abs(np.var(input_vector, axis=0) - 1) < 1e-3))
_check_zero_mean_unit_variance(input_features[0, : fbank_feat_lengths[0]])
_check_zero_mean_unit_variance(input_features[1, : fbank_feat_lengths[1]])
_check_zero_mean_unit_variance(input_features[2])
# make sure that if max_length < longest -> then pad to max_length
self.assertEqual(input_features.shape, (3, 4, 24))
speech_inputs = [floats_list((1, x))[0] for x in range(800, 1400, 200)]
inputs = feature_extractor(
speech_inputs,
padding="longest",
max_length=16,
truncation=True,
return_tensors="np",
return_attention_mask=True,
)
input_features = inputs.input_features
attention_mask = inputs.attention_mask
fbank_feat_lengths = np.sum(attention_mask == 1, axis=1)
_check_zero_mean_unit_variance(input_features[0, : fbank_feat_lengths[0]])
_check_zero_mean_unit_variance(input_features[1, : fbank_feat_lengths[1]])
_check_zero_mean_unit_variance(input_features[2])
# make sure that if max_length < longest -> then pad to max_length
self.assertEqual(input_features.shape, (3, 6, 24))