Update chatterbox/src/chatterbox/tts.py

chatterbox/src/chatterbox/tts.py

@@ -45,10 +45,10 @@ def punc_norm(text: str) -> str:
         ("—", "-"),
         ("–", "-"),
         (" ,", ","),
-        ("“", "\""),
-        ("”", "\""),
-        ("‘", "'"),
-        ("’", "'"),
+        (""", "\""),
+        (""", "\""),
+        ("'", "'"),
+        ("'", "'"),
     ]
     for old_char_sequence, new_char in punc_to_replace:
         text = text.replace(old_char_sequence, new_char)
@@ -131,12 +131,12 @@ class ChatterboxTTS:
 
         ve = VoiceEncoder()
         ve.load_state_dict(
-            torch.load(ckpt_dir / "ve.pt")
+            torch.load(ckpt_dir / "ve.pt", map_location=device)
         )
         ve.to(device).eval()
 
         t3 = T3()
-        t3_state = torch.load(ckpt_dir / "t3_cfg.pt")
+        t3_state = torch.load(ckpt_dir / "t3_cfg.pt", map_location=device)
         if "model" in t3_state.keys():
             t3_state = t3_state["model"][0]
         t3.load_state_dict(t3_state)
@@ -144,7 +144,7 @@ class ChatterboxTTS:
 
         s3gen = S3Gen()
         s3gen.load_state_dict(
-            torch.load(ckpt_dir / "s3gen.pt")
+            torch.load(ckpt_dir / "s3gen.pt", map_location=device)
         )
         s3gen.to(device).eval()
 
@@ -154,7 +154,7 @@ class ChatterboxTTS:
 
         conds = None
         if (builtin_voice := ckpt_dir / "conds.pt").exists():
-            conds = Conditionals.load(builtin_voice).to(device)
+            conds = Conditionals.load(builtin_voice, map_location=device).to(device)
 
         return cls(t3, s3gen, ve, tokenizer, device, conds=conds)
 
@@ -267,4 +267,4 @@ class ChatterboxTTS:
             )
             wav = wav.squeeze(0).detach().cpu().numpy()
             watermarked_wav = self.watermarker.apply_watermark(wav, sample_rate=self.sr)
-        return torch.from_numpy(watermarked_wav).unsqueeze(0)
+        return torch.from_numpy(watermarked_wav).unsqueeze(0)