Language model weights can be specified on cli

speecht-cli

@@ -77,6 +77,12 @@ class CLI:
                              'Specify a directory containing `kenlm-model.binary`, '
                              '`vocabulary` and `trie`. '
                              'Language model must be binary format with probing hash table.')
+    parser.add_argument('--lm-weight', dest='lm_weight', type=float, default=0.8,
+                        help='The weight multiplied with the language model score')
+    parser.add_argument('--word-count-weight', dest='word_count_weight', type=float, default=0.0,
+                        help='The weight added for each word')
+    parser.add_argument('--valid-word-count-weight', dest='valid_word_count_weight', type=float, default=2.3,
+                        help='The weight added for each in vocabulary word')
 
   def _add_evaluation_parser(self):
     evaluation_parser = self.subparsers.add_parser('evaluate', help='Evaluate the development or test set.',

speecht/speech_model.py

@@ -81,17 +81,22 @@ def add_training_ops(self, learning_rate: bool = 1e-3, learning_rate_decay_facto
         self.update = optimizer.apply_gradients(zip(clipped_gradients, trainables),
                                                 global_step=self.global_step, name='apply_gradients')
 
-  def add_decoding_ops(self, language_model: str = None):
+  def add_decoding_ops(self, language_model: str = None, lm_weight: float = 0.8, word_count_weight: float = 0.0,
+                       valid_word_count_weight: float = 2.3):
     """
     Add the ops for decoding
-
+j
     Args:
       language_model: the file path to the language model to use for beam search decoding or None
+      word_count_weight: The weight added for each added word
+      valid_word_count_weight: The weight added for each in vocabulary word
+      lm_weight: The weight multiplied with the language model scoring
     """
     with tf.name_scope('decoding'):
-      self.lm_weight = tf.placeholder_with_default(0.8, shape=(), name='language_model_weight')
-      self.word_count_weight = tf.placeholder_with_default(0.0, shape=(), name='word_count_weight')
-      self.valid_word_count_weight = tf.placeholder_with_default(2.3, shape=(), name='valid_word_count_weight')
+      self.lm_weight = tf.placeholder_with_default(lm_weight, shape=(), name='language_model_weight')
+      self.word_count_weight = tf.placeholder_with_default(word_count_weight, shape=(), name='word_count_weight')
+      self.valid_word_count_weight = tf.placeholder_with_default(valid_word_count_weight, shape=(),
+                                                                 name='valid_word_count_weight')
 
       if language_model:
         self.softmaxed = tf.log(tf.nn.softmax(self.logits, name='softmax') + 1e-8) / math.log(10)
@@ -304,7 +309,10 @@ def create_default_model(flags, input_size: int, speech_input: BaseInputLoader)
     model.add_decoding_ops()
   else:
     model.add_training_ops()
-    model.add_decoding_ops(language_model=flags.language_model)
+    model.add_decoding_ops(language_model=flags.language_model,
+                           lm_weight=flags.lm_weight,
+                           word_count_weight=flags.word_count_weight,
+                           valid_word_count_weight=flags.valid_word_count_weight)
 
   model.finalize(log_dir=flags.log_dir,
                  run_name=flags.run_name,