[Cherry-pick 2.0.1] Fix virtual function issue with CTC decoder (#3230)

test/torchaudio_unittest/models/decoder/ctc_decoder_test.py

@@ -169,3 +169,19 @@ def test_index_to_tokens(self, tokens):
 
         expected_tokens = ["|", "f", "|", "o", "a"]
         self.assertEqual(tokens, expected_tokens)
+
+    def test_lm_lifecycle(self):
+        """Passing lm without assiging it to a vaiable won't cause runtime error
+
+        https://github.com/pytorch/audio/issues/3218
+        """
+        from torchaudio.models.decoder import ctc_decoder
+
+        from .ctc_decoder_utils import CustomZeroLM
+
+        decoder = ctc_decoder(
+            lexicon=get_asset_path("decoder/lexicon.txt"),
+            tokens=get_asset_path("decoder/tokens.txt"),
+            lm=CustomZeroLM(),
+        )
+        decoder(torch.zeros((1, 3, NUM_TOKENS), dtype=torch.float32))

torchaudio/models/decoder/_ctc_decoder.py

@@ -269,6 +269,12 @@ def __init__(
             )
         else:
             self.decoder = _LexiconFreeDecoder(decoder_options, lm, silence, self.blank, transitions)
+        # https://github.com/pytorch/audio/issues/3218
+        # If lm is passed like rvalue reference, the lm object gets garbage collected,
+        # and later call to the lm fails.
+        # This ensures that lm object is not deleted as long as the decoder is alive.
+        # https://github.com/pybind/pybind11/discussions/4013
+        self.lm = lm
 
     def _get_tokens(self, idxs: torch.IntTensor) -> torch.LongTensor:
         idxs = (g[0] for g in it.groupby(idxs))