Create norm.py

norm.py

@@ -0,0 +1,17 @@
+import neologdn
+import MeCab
+import re
+from transformers.models.whisper.english_normalizer import BasicTextNormalizer
+
+wakati = MeCab.Tagger("-Owakati")
+special_characters = '[\,\、\。\．\「\」\…\？\・\!\-\;\:\"\“\%\‘\”\�]'
+
+
+def norm_everything(batch): 
+    batch["sentence"] = neologdn.normalize(batch["sentence"]).strip() 
+    batch["sentence"] = normalizer(batch["sentence"]).strip() 
+    batch["sentence"] = wakati.parse(batch["sentence"]).strip() 
+    batch["sentence"] = re.sub(special_characters,'', batch["sentence"]).strip() 
+    return batch 
+
+ds = ds.map(norm_everything)