Change to g2p_en

inference_cli.py

@@ -2,6 +2,7 @@
 
 from typing import Any, Callable
 import json
+import os
 
 import fire
 import torch
@@ -146,18 +147,19 @@ class InferenceCLI:
         output_path: str = "./output.wav",
     ):
 
-        from montreal_forced_aligner.g2p.generator import PyniniConsoleGenerator
+        from g2p_en import G2p
+        import nltk
 
         self.init_speaker_model()
 
         if not self.g2p:
-            self.g2p = PyniniConsoleGenerator(
-                g2p_model_path=self.model.g2p_model_path,
-                strict_graphemes=False,
-                num_pronunciations=1,
-                include_bracketed=False
-            )
-            self.g2p.setup()
+            if not os.path.exists(
+                os.path.expanduser(
+                    "~/nltk_data/taggers/averaged_perceptron_tagger_eng"
+                )
+            ):
+                nltk.download("averaged_perceptron_tagger_eng")
+            self.g2p = G2p()
 
         if not self.word2phone:
             self.word2phone = json.load(

utils/phonemize.py

@@ -4,12 +4,10 @@ import re
 def g2p_resolve(word, g2p_model):
     """Call G2P to generate pronunciation (used for handling OOV words)."""
     try:
-        result = g2p_model.rewriter(word.lower())
-        if result and result[0][0]:
-            return result[0][0].split()
+        result = g2p_model(word)
+        return result
     except Exception:
         return None
-    return None
 
 
 def text_norm(s):