fix the bugs

examples/speech_to_text/prep_audio_data.py

@@ -185,7 +185,7 @@ class AudioDataset(Dataset):
 
         if need_waveform:
             offset = item.get('offset', False)
-            if offset is not False:
+            if offset:
                 waveform, sample_rate = torchaudio.load(audio,
                                                         frame_offset=offset,
                                                         num_frames=item["n_frames"])
@@ -331,7 +331,7 @@ def process(args):
                     audio_path = item["audio"]
 
                     # add offset and frames info
-                    if item.get("offset", False) is not False:
+                    if item.get("offset", False):
                         audio_path = f"{audio_path}:{item['offset']}:{n_frames}"
                     manifest["audio"].append(audio_path)
                 else:

fairseq/criterions/label_smoothed_cross_entropy_with_ctc.py

@@ -37,13 +37,13 @@ class LabelSmoothedCrossEntropyCriterionWithCTC(
         """Add criterion-specific arguments to the parser."""
         LabelSmoothedCrossEntropyCriterion.add_args(parser)
         CtcCriterion.add_args(parser)
-        parser.add_argument(
-            "--ctc-weight",
-            default=0.0,
-            type=float,
-            metavar="D",
-            help="weight of CTC loss",
-        )
+        # parser.add_argument(
+        #     "--ctc-weight",
+        #     default=0.0,
+        #     type=float,
+        #     metavar="D",
+        #     help="weight of CTC loss",
+        # )
 
     def forward(self, model, sample, reduce=True):
         """Compute the loss for the given sample.