Fix language assignment

TTS/config/shared_configs.py

@@ -193,9 +193,12 @@ class BaseDatasetConfig(Coqpit):
             Name of the dataset meta file. Or a list of speakers to be ignored at training for multi-speaker datasets.
             Defaults to None.
 
-        unused_speakers (List):
+        ignored_speakers (List):
             List of speakers IDs that are not used at the training. Default None.
 
+        language (str):
+            Language code of the dataset. If defined, it overrides `phoneme_language`. Defaults to None.
+
         meta_file_val (str):
             Name of the dataset meta file that defines the instances used at validation.
 

TTS/tts/datasets/dataset.py

@@ -273,9 +273,9 @@ class TTSDataset(Dataset):
         item = args[0]
         func_args = args[1]
         text, wav_file, *_ = item
-        if item[3] is not None and item[3] != '':
-            # If language is specified in dataset, overwrite phoneme_language key
-            func_args[3] = item[3]
+        func_args[3] = (
+            item[3] if item[3] else func_args[3]
+        )  # override phoneme language if specified by the dataset formatter
         phonemes = TTSDataset._load_or_generate_phoneme_sequence(wav_file, text, *func_args)
         return phonemes
 

recipes/vctk/vits/train_vits.py

@@ -10,7 +10,9 @@ from TTS.tts.utils.speakers import SpeakerManager
 from TTS.utils.audio import AudioProcessor
 
 output_path = os.path.dirname(os.path.abspath(__file__))
-dataset_config = BaseDatasetConfig(name="vctk", meta_file_train="", path=os.path.join(output_path, "../VCTK/"))
+dataset_config = BaseDatasetConfig(
+    name="vctk", meta_file_train="", language="en-us", path=os.path.join(output_path, "../VCTK/")
+)
 
 
 audio_config = BaseAudioConfig(
@@ -49,7 +51,6 @@ config = VitsConfig(
     epochs=1000,
     text_cleaner="english_cleaners",
     use_phonemes=True,
-    phoneme_language="en-us",
     phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
     compute_input_seq_cache=True,
     print_step=25,