norm for en

Signed-off-by: nithinraok <[email protected]>

Author: nithinraok

nemo_asr/run_eval_ml.py

@@ -166,9 +166,13 @@ def download_audio_files(batch):
     if isinstance(transcriptions, tuple) and len(transcriptions) == 2:
         transcriptions = transcriptions[0]
 
-    references = all_data["references"]
-    references = [data_utils.ml_normalizer(ref) for ref in references]
-    predictions = [data_utils.ml_normalizer(pred.text) for pred in transcriptions]
+    references = all_data["references"] 
+    if LANGUAGE == "en": # English is handled by the English normalizer
+        references = [data_utils.normalizer(ref) for ref in references]
+        predictions = [data_utils.normalizer(pred.text) for pred in transcriptions]
+    else:
+        references = [data_utils.ml_normalizer(ref) for ref in references]
+        predictions = [data_utils.ml_normalizer(pred.text) for pred in transcriptions]
 
     avg_time = total_time / len(all_data["audio_filepaths"])
 

nemo_asr/run_nemo_ml.sh

@@ -7,7 +7,7 @@
 export PYTHONPATH="..":$PYTHONPATH
 
 # Configuration
-MODEL_ID="nvidia/canary-1b-v2"
+MODEL_ID="nvidia/parakeet-tdt-0.6b-v3"  #"nvidia/canary-1b-v2"
 
 BATCH_SIZE=64