diff --git a/docs/source/index.rst b/docs/source/index.rst
index 596a636b1b..49ffb9af22 100644
--- a/docs/source/index.rst
+++ b/docs/source/index.rst
@@ -39,6 +39,7 @@ The :mod:`torchaudio` package consists of I/O, popular datasets and common audio
    compliance.kaldi
    kaldi_io
    utils
+   transducer
    
    
 .. toctree::
diff --git a/docs/source/transducer.rst b/docs/source/transducer.rst
new file mode 100644
index 0000000000..505ab0187c
--- /dev/null
+++ b/docs/source/transducer.rst
@@ -0,0 +1,23 @@
+.. role:: hidden
+    :class: hidden-section
+
+torchaudio.prototype.transducer
+===============================
+
+.. currentmodule:: torchaudio.prototype.transducer
+
+.. note::
+
+    The RNN transducer loss is a prototype feature, see `here <https://pytorch.org/audio>`_ to learn more about the nomenclature. It is only available within the nightlies, and also needs to be imported explicitly using: :code:`from torchaudio.prototype.transducer import rnnt_loss, RNNTLoss`.
+
+rnnt_loss
+---------
+
+.. autofunction:: rnnt_loss
+
+:hidden:`RNNTLoss`
+~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: RNNTLoss
+
+  .. automethod:: forward