Merge pull request #83 from TensorSpeech/fix/tflite

Fix TFLite Conversion for Transducer Greedy

Author: nglehuy

examples/conformer/tflite_subword_conformer.py

@@ -64,9 +64,9 @@
 
 concrete_func = conformer.make_tflite_function(greedy=True).get_concrete_function()
 converter = tf.lite.TFLiteConverter.from_concrete_functions([concrete_func])
+converter.experimental_new_converter = True
 converter.optimizations = [tf.lite.Optimize.DEFAULT]
-converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS,
-                                       tf.lite.OpsSet.SELECT_TF_OPS]
+converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS, tf.lite.OpsSet.SELECT_TF_OPS]
 tflite_model = converter.convert()
 
 if not os.path.exists(os.path.dirname(args.output)):

setup.py

@@ -37,7 +37,7 @@
 
 setuptools.setup(
     name="TensorFlowASR",
-    version="0.5.1",
+    version="0.5.2",
     author="Huy Le Nguyen",
     author_email="[email protected]",
     description="Almost State-of-the-art Automatic Speech Recognition using Tensorflow 2",

tensorflow_asr/models/transducer.py

@@ -417,73 +417,51 @@ def __perform_greedy(self,
         with tf.name_scope(f"{self.name}_greedy"):
             time = tf.constant(0, dtype=tf.int32)
             total = encoded_length
-            # Initialize prediction with a blank
-            # Prediction can not be longer than the encoded of audio plus blank
-            prediction = tf.TensorArray(
-                dtype=tf.int32,
-                size=(total + 1),
-                dynamic_size=False,
-                element_shape=tf.TensorShape([]),
-                clear_after_read=False
-            )
 
             hypothesis = Hypothesis(
                 index=tf.constant(0, dtype=tf.int32),
-                prediction=prediction.write(0, predicted),
+                prediction=tf.ones([total + 1], dtype=tf.int32) * self.text_featurizer.blank,
                 states=states
             )
 
             def condition(time, total, encoded, hypothesis): return tf.less(time, total)
 
             def body(time, total, encoded, hypothesis):
+                predicted = tf.gather_nd(hypothesis.prediction, tf.expand_dims(hypothesis.index, axis=-1))
+
                 ytu, new_states = self.decoder_inference(
                     # avoid using [index] in tflite
                     encoded=tf.gather_nd(encoded, tf.expand_dims(time, axis=-1)),
-                    predicted=hypothesis.prediction.read(hypothesis.index),
+                    predicted=predicted,
                     states=hypothesis.states
                 )
-                char = tf.argmax(ytu, axis=-1, output_type=tf.int32)  # => argmax []
-
-                index, char, new_states = tf.cond(
-                    tf.equal(char, self.text_featurizer.blank),
-                    true_fn=lambda: (
-                        hypothesis.index,
-                        hypothesis.prediction.read(hypothesis.index),
-                        hypothesis.states
-                    ),
-                    false_fn=lambda: (
-                        hypothesis.index + 1,
-                        char,
-                        new_states
-                    )
+                new_predicted = tf.argmax(ytu, axis=-1, output_type=tf.int32)  # => argmax []
+
+                index, new_predicted, new_states = tf.cond(
+                    tf.equal(new_predicted, self.text_featurizer.blank),
+                    true_fn=lambda: (hypothesis.index, predicted, hypothesis.states),
+                    false_fn=lambda: (hypothesis.index + 1, new_predicted, new_states)
                 )
 
                 hypothesis = Hypothesis(
                     index=index,
-                    prediction=hypothesis.prediction.write(index, char),
+                    prediction=tf.tensor_scatter_nd_update(
+                        hypothesis.prediction,
+                        indices=tf.reshape(index, [1, 1]),
+                        updates=tf.expand_dims(new_predicted, axis=-1)
+                    ),
                     states=new_states
                 )
 
                 return time + 1, total, encoded, hypothesis
 
             time, total, encoded, hypothesis = tf.while_loop(
-                condition,
-                body,
+                condition, body,
                 loop_vars=(time, total, encoded, hypothesis),
                 parallel_iterations=parallel_iterations,
                 swap_memory=swap_memory
             )
 
-            # Gather predicted sequence
-            hypothesis = Hypothesis(
-                index=hypothesis.index,
-                prediction=tf.gather_nd(
-                    params=hypothesis.prediction.stack(),
-                    indices=tf.expand_dims(tf.range(hypothesis.index + 1), axis=-1)
-                ),
-                states=hypothesis.states
-            )
-
             return hypothesis
 
     # -------------------------------- BEAM SEARCH -------------------------------------