Fixed sampler override and audio_key in prepare_audio_data (NVIDIA#10980

) Signed-off-by: Ante Jukić <ajukic@nvidia.com>
minitu · Oct 22, 2024 · 70d8cc1 · 70d8cc1
1 parent 746203a
commit 70d8cc1
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 3 deletions.
diff --git a/examples/audio/process_audio.py b/examples/audio/process_audio.py
@@ -159,8 +159,8 @@ def main(cfg: ProcessConfig) -> ProcessConfig:
     audio_to_audio_model.set_trainer(trainer)
     audio_to_audio_model = audio_to_audio_model.eval()
 
-    # override sampler
-    if cfg.sampler is not None:
+    # override sampler if necessary
+    if cfg.sampler:
         logging.info('Overriding sampler with %s', cfg.sampler)
 
         if hasattr(audio_to_audio_model, 'sampler'):

diff --git a/nemo/collections/asr/parts/utils/transcribe_utils.py b/nemo/collections/asr/parts/utils/transcribe_utils.py
@@ -314,7 +314,7 @@ def prepare_audio_data(cfg: DictConfig) -> Tuple[List[str], bool]:
         with NamedTemporaryFile(mode='w', suffix='.json', delete=False) as f:
             for item in read_and_maybe_sort_manifest(cfg.dataset_manifest, try_sort=cfg.presort_manifest):
                 audio_file = get_full_path(audio_file=item[audio_key], manifest_file=cfg.dataset_manifest)
-                item[audio_key] = audio_file
+                item['audio_filepath'] = audio_file
                 filepaths.append(audio_file)
                 f.write(json.dumps(item) + "\n")
         sorted_manifest_path = f.name