✍️ update base runner and dataset for empty eval dataset

Author: nglehuy

examples/conformer/config.yml

@@ -24,7 +24,7 @@ speech_config:
   normalize_per_feature: False
 
 decoder_config:
-  vocabulary: ./vocabularies/librispeech/librispeech_train_4_1030.subwords
+  vocabulary: null
   target_vocab_size: 4096
   max_subword_length: 4
   blank_at_zero: True
@@ -74,7 +74,7 @@ learning_config:
           num_masks: 1
           mask_factor: 27
     data_paths:
-      - /media/nlhuy/Data/ML/Datasets/ASR/Raw/LibriSpeech/train-clean-100/transcripts.tsv
+      - /mnt/Data/ML/Datasets/ASR/Raw/LibriSpeech/train-clean-100/transcripts.tsv
     tfrecords_dir: /mnt/Miscellanea/Datasets/Speech/LibriSpeech/tfrecords
     shuffle: True
     cache: True

setup.py

@@ -22,7 +22,7 @@
 
 setuptools.setup(
     name="TensorFlowASR",
-    version="0.8.0",
+    version="0.8.1",
     author="Huy Le Nguyen",
     author_email="[email protected]",
     description="Almost State-of-the-art Automatic Speech Recognition using Tensorflow 2",

tensorflow_asr/datasets/base_dataset.py

@@ -36,7 +36,7 @@ def __init__(self,
                  use_tf: bool = False,
                  stage: str = "train",
                  **kwargs):
-        self.data_paths = data_paths
+        self.data_paths = data_paths or []
         self.augmentations = augmentations  # apply augmentation
         self.cache = cache  # whether to cache WHOLE transformed dataset to memory
         self.shuffle = shuffle  # whether to shuffle tf.data.Dataset

tensorflow_asr/runners/base_runners.py

@@ -136,7 +136,10 @@ def set_eval_data_loader(self, eval_dataset, eval_bs=None):
             return
         if not eval_bs: eval_bs = self.config.batch_size
         self.eval_data = eval_dataset.create(eval_bs * self.strategy.num_replicas_in_sync)
-        self.eval_data_loader = self.strategy.experimental_distribute_dataset(self.eval_data)
+        if self.eval_data is None:
+            self.eval_data_loader = None
+        else:
+            self.eval_data_loader = self.strategy.experimental_distribute_dataset(self.eval_data)
         self.eval_steps_per_epoch = eval_dataset.total_steps
 
     # -------------------------------- CHECKPOINTS -------------------------------------