Merge pull request #53 from TensorSpeech/dev/refactor

Add additional field in config, remove required tensorflow dependency

Author: nglehuy

README.md

@@ -67,6 +67,8 @@ TensorFlowASR implements some automatic speech recognition architectures such as
 
 ## Installation
 
+Install `tensorflow>=2.3.0` or `tf-nightly`.
+
 For training and testing, you should use `git clone` for installing necessary packages from other authors (`ctc_decoders`, `rnnt_loss`, etc.)
 
 ### Installing via PyPi

examples/streaming_transducer/config.yml

@@ -80,7 +80,7 @@ learning_config:
     batch_size: 2
     accumulation_steps: 1
     num_epochs: 20
-    outdir: /mnt/SpeechProcessing/Trained/local/streaming_transducer
+    outdir: /mnt/d/SpeechProcessing/Trained/local/streaming_transducer
     log_interval_steps: 300
     eval_interval_steps: 500
     save_interval_steps: 1000

setup.py

@@ -18,7 +18,6 @@
     long_description = fh.read()
 
 requirements = [
-    "tensorflow>=2.3.0",
     "tensorflow-datasets>=3.2.1,<4.0.0",
     "tensorflow-addons>=0.10.0",
     "setuptools>=47.1.1",
@@ -38,7 +37,7 @@
 
 setuptools.setup(
     name="TensorFlowASR",
-    version="0.3.0",
+    version="0.3.1",
     author="Huy Le Nguyen",
     author_email="[email protected]",
     description="Almost State-of-the-art Automatic Speech Recognition using Tensorflow 2",

tensorflow_asr/configs/config.py

@@ -24,6 +24,7 @@ def __init__(self, config: dict = None):
         self.eval_paths = config.get("eval_paths", None)
         self.test_paths = config.get("test_paths", None)
         self.tfrecords_dir = config.get("tfrecords_dir", None)
+        self.additional_properties = config.get("additional_properties", {})
 
 
 class RunningConfig:
@@ -36,6 +37,7 @@ def __init__(self, config: dict = None):
         self.log_interval_steps = config.get("log_interval_steps", 500)
         self.save_interval_steps = config.get("save_interval_steps", 500)
         self.eval_interval_steps = config.get("eval_interval_steps", 1000)
+        self.additional_properties = config.get("additional_properties", {})
 
 
 class LearningConfig:
@@ -45,6 +47,7 @@ def __init__(self, config: dict = None):
         self.dataset_config = DatasetConfig(config.get("dataset_config"))
         self.optimizer_config = config.get("optimizer_config", {})
         self.running_config = RunningConfig(config.get("running_config"))
+        self.additional_properties = config.get("additional_properties", {})
 
 
 class Config:
@@ -55,5 +58,6 @@ def __init__(self, path: str, learning: bool):
         self.speech_config = config.get("speech_config", {})
         self.decoder_config = config.get("decoder_config", {})
         self.model_config = config.get("model_config", {})
+        self.additional_properties = config.get("additional_properties", {})
         if learning:
             self.learning_config = LearningConfig(config.get("learning_config"))

tensorflow_asr/models/layers/multihead_attention.py

@@ -117,7 +117,7 @@ def call_qkv(self, query, key, value, training=False):
     def call_attention(self, query, key, value, logits, training=False, mask=None):
         if mask is not None:
             if len(mask.shape) < 2:
-                raise ValueError("'mask' must have atleast 2 dimensions")
+                raise ValueError("'mask' must have at least 2 dimensions")
             if query.shape[-2] != mask.shape[-2]:
                 raise ValueError(
                     "mask's second to last dimension must be equal to "