def __init__(self, dataset_config):
"""Initialize the DeepSpeechDataset class.
Args:
dataset_config: DatasetConfig object.
"""
self.config = dataset_config
# Instantiate audio feature extractor.
self.audio_featurizer = featurizer.AudioFeaturizer(
sample_rate=self.config.audio_config.sample_rate,
window_ms=self.config.audio_config.window_ms,
stride_ms=self.config.audio_config.stride_ms)
# Instantiate text feature extractor.
self.text_featurizer = featurizer.TextFeaturizer(
vocab_file=self.config.vocab_file_path)
self.speech_labels = self.text_featurizer.speech_labels
self.entries = _preprocess_data(self.config.data_path)
# The generated spectrogram will have 161 feature bins.
self.num_feature_bins = 161
def __init__(self, dataset_config):
"""Initialize the DeepSpeechDataset class.
Args:
dataset_config: DatasetConfig object.
"""
self.config = dataset_config
# Instantiate audio feature extractor.
self.audio_featurizer = featurizer.AudioFeaturizer(
sample_rate=self.config.audio_config.sample_rate,
frame_length=self.config.audio_config.frame_length,
frame_step=self.config.audio_config.frame_step,
fft_length=self.config.audio_config.fft_length)
# Instantiate text feature extractor.
self.text_featurizer = featurizer.TextFeaturizer(
vocab_file=self.config.vocab_file_path)
self.speech_labels = self.text_featurizer.speech_labels
self.features, self.labels = _preprocess_data(
self.config, self.audio_featurizer,
self.text_featurizer.token_to_idx)
self.num_feature_bins = (self.features[0].shape[1]
if len(self.features) else None)