diff --git a/deep_speech_2/model.py b/deep_speech_2/model.py
index c8766deb1d33a12b9e96f00696847fd000c2b78d..c2e440b3ae4a8455d41de7b293e95123b7038931 100644
--- a/deep_speech_2/model.py
+++ b/deep_speech_2/model.py
@@ -120,6 +120,16 @@ class DeepSpeech2Model(object):
             feeding=feeding_dict)
 
     def infer_loss_batch(self, infer_data):
+        """Model inference. Infer the ctc loss for a batch of speech
+        utterances.
+
+        :param infer_data: List of utterances to infer, with each utterance a
+                           tuple of audio features and transcription text (empty
+                           string).
+        :type infer_data: list
+        :return: List of ctc loss.
+        :rtype: List of float
+        """
         # define inferer
         if self._loss_inferer == None:
             self._loss_inferer = paddle.inference.Inference(