fix

deepspeech/frontend/augmentor/shift_perturb.py

@@ -32,7 +32,8 @@ class ShiftPerturbAugmentor(AugmentorBase):
         self._rng = rng
 
     def randomize_parameters(self):
-        self.shift_ms = self._rng.uniform(self._min_shift_ms, self._max_shift_ms)
+        self.shift_ms = self._rng.uniform(self._min_shift_ms,
+                                          self._max_shift_ms)
 
     def apply(self, audio_segment):
         audio_segment.shift(self.shift_ms)
@@ -49,7 +50,6 @@ class ShiftPerturbAugmentor(AugmentorBase):
     #         self.randomize_parameters()
     #     self.apply(audio_segment)
 
-
     # def transform_audio(self, audio_segment):
     #     """Shift audio.
 
@@ -60,5 +60,3 @@ class ShiftPerturbAugmentor(AugmentorBase):
     #     """
     #     shift_ms = self._rng.uniform(self._min_shift_ms, self._max_shift_ms)
     #     audio_segment.shift(shift_ms)
-
-

deepspeech/frontend/augmentor/spec_augment.py

@@ -174,7 +174,6 @@ class SpecAugmentor(AugmentorBase):
             assert t_0 <= t_0 + t
         return xs
 
-
     # def mask_freq(self, xs, replace_with_zero=False):
     #     n_bins = xs.shape[0]
     #     for i in range(0, self.n_freq_masks):
@@ -208,7 +207,6 @@ class SpecAugmentor(AugmentorBase):
     #         self._time_mask = (t_0, t_0 + t)
     #     return xs
 
-
     # def transform_feature(self, xs: np.ndarray, single=True):
     #     """
     #     Args:

deepspeech/frontend/augmentor/speed_perturb.py

@@ -79,7 +79,6 @@ class SpeedPerturbAugmentor(AugmentorBase):
             self._rates = np.linspace(
                 self._min_rate, self._max_rate, self._num_rates, endpoint=True)
 
-
     def randomize_parameters(self):
         if self._num_rates < 0:
             self.speed_rate = self._rng.uniform(self._min_rate, self._max_rate)

deepspeech/io/collator.py

@@ -174,7 +174,6 @@ class SpeechCollator():
         self._stride_ms = stride_ms
         self._target_sample_rate = target_sample_rate
 
-
         self._speech_featurizer = SpeechFeaturizer(
             unit_type=unit_type,
             vocab_filepath=vocab_filepath,
@@ -231,7 +230,8 @@ class SpeechCollator():
         self._augmentation_pipeline.randomize_parameters_audio_transform()
 
     def randomize_feature_parameters(self, n_frames, n_bins):
-        self._augmentation_pipeline.randomize_parameters_feature_transform(n_frames, n_bins)
+        self._augmentation_pipeline.randomize_parameters_feature_transform(
+            n_frames, n_bins)
 
     def process_feature_and_transform(self, audio_file, transcript):
         """Load, augment, featurize and normalize for speech data.
@@ -261,7 +261,6 @@ class SpeechCollator():
 
         return specgram, transcript_part
 
-
     # def process_utterance(self, audio_file, transcript, single=True):
     #     """Load, augment, featurize and normalize for speech data.
 
@@ -282,7 +281,6 @@ class SpeechCollator():
     #     # audio augment
     #     self._augmentation_pipeline.transform_audio(speech_segment)
 
-
     #     # Spectrum transform
     #     specgram, transcript_part = self._speech_featurizer.featurize(
     #         speech_segment, self._keep_transcription_text)
@@ -357,7 +355,9 @@ class SpeechCollator():
         for i in range(len(padded_audios)):
             if not self._randomize_each_batch:
                 self.randomize_feature_parameters(audio_lens[i], n_bins)
-            padded_audios[i] = self._augmentation_pipeline.apply_feature_transform(padded_audios[i])
+            padded_audios[
+                i] = self._augmentation_pipeline.apply_feature_transform(
+                    padded_audios[i])
 
         return utts, padded_audios, audio_lens, padded_texts, text_lens