Fix phonemes

TTS/bin/find_unique_phonemes.py

@@ -7,7 +7,7 @@ from tqdm.contrib.concurrent import process_map
 
 from TTS.config import load_config
 from TTS.tts.datasets import load_tts_samples
-from TTS.tts.utils.text import text2phone
+from TTS.tts.utils.text import text2phone, phoneme_to_sequence
 
 
 def compute_phonemes(item):

TTS/tts/datasets/dataset.py

@@ -273,7 +273,7 @@ class TTSDataset(Dataset):
         item = args[0]
         func_args = args[1]
         text, wav_file, *_ = item
-        func_args[3] = item[4]
+        func_args[3] = item[3]
         phonemes = TTSDataset._load_or_generate_phoneme_sequence(wav_file, text, *func_args)
         return phonemes
 

TTS/tts/models/vits.py

@@ -543,6 +543,7 @@ class Vits(BaseTTS):
             "style_wav": style_wav,
             "d_vector": d_vector,
             "language_id": language_id,
+            "language_name": language_name,
         }
 
     def forward(
@@ -1061,6 +1062,7 @@ class Vits(BaseTTS):
                     d_vector=aux_inputs["d_vector"],
                     style_wav=aux_inputs["style_wav"],
                     language_id=aux_inputs["language_id"],
+                    language_name=aux_inputs["language_name"],
                     enable_eos_bos_chars=self.config.enable_eos_bos_chars,
                     use_griffin_lim=True,
                     do_trim_silence=False,

TTS/tts/utils/synthesis.py

@@ -15,7 +15,7 @@ if "tensorflow" in installed or "tensorflow-gpu" in installed:
     import tensorflow as tf
 
 
-def text_to_seq(text, CONFIG, custom_symbols=None):
+def text_to_seq(text, CONFIG, custom_symbols=None, language=None):
     text_cleaner = [CONFIG.text_cleaner]
     # text ot phonemes to sequence vector
     if CONFIG.use_phonemes:
@@ -23,7 +23,7 @@ def text_to_seq(text, CONFIG, custom_symbols=None):
             phoneme_to_sequence(
                 text,
                 text_cleaner,
-                CONFIG.phoneme_language,
+                language if language else CONFIG.phoneme_language,
                 CONFIG.enable_eos_bos_chars,
                 tp=CONFIG.characters,
                 add_blank=CONFIG.add_blank,
@@ -212,6 +212,7 @@ def synthesis(
     do_trim_silence=False,
     d_vector=None,
     language_id=None,
+    language_name=None,
     backend="torch",
 ):
     """Synthesize voice for the given text using Griffin-Lim vocoder or just compute output features to be passed to
@@ -262,7 +263,7 @@ def synthesis(
     if hasattr(model, "make_symbols"):
         custom_symbols = model.make_symbols(CONFIG)
     # preprocess the given text
-    text_inputs = text_to_seq(text, CONFIG, custom_symbols=custom_symbols)
+    text_inputs = text_to_seq(text, CONFIG, custom_symbols=custom_symbols, language=language_name)
     # pass tensors to backend
     if backend == "torch":
         if speaker_id is not None: