From d39878eac0c83c251f68c9b2e7539be3e88fb4e3 Mon Sep 17 00:00:00 2001 From: Eren G??lge Date: Wed, 24 May 2023 11:55:48 +0200 Subject: [PATCH] Fixup --- TTS/tts/models/vits.py | 8 ++++++-- TTS/tts/utils/fairseq.py | 2 +- 2 files changed, 7 insertions(+), 3 deletions(-) diff --git a/TTS/tts/models/vits.py b/TTS/tts/models/vits.py index 7240de36..6770a98e 100644 --- a/TTS/tts/models/vits.py +++ b/TTS/tts/models/vits.py @@ -1724,7 +1724,9 @@ class Vits(BaseTTS): self.eval() assert not self.training - def load_fairseq_checkpoint(self, config, checkpoint_dir, eval=False): # pylint: disable=unused-argument, redefined-builtin + def load_fairseq_checkpoint( + self, config, checkpoint_dir, eval=False + ): # pylint: disable=unused-argument, redefined-builtin """Load VITS checkpoints released by fairseq here: https://github.com/facebookresearch/fairseq/tree/main/examples/mms Performs some changes for compatibility. @@ -1735,6 +1737,8 @@ class Vits(BaseTTS): """ import json + from TTS.tts.utils.text.cleaners import basic_cleaners + self.disc = None # set paths config_file = os.path.join(checkpoint_dir, "config.json") @@ -1751,7 +1755,7 @@ class Vits(BaseTTS): self.text_encoder.emb = nn.Embedding(vocab.num_chars, config.model_args.hidden_channels) self.tokenizer = TTSTokenizer( use_phonemes=False, - text_cleaner=None, + text_cleaner=basic_cleaners, characters=vocab, phonemizer=None, add_blank=config_org["data"]["add_blank"], diff --git a/TTS/tts/utils/fairseq.py b/TTS/tts/utils/fairseq.py index b1012958..3d8eec2b 100644 --- a/TTS/tts/utils/fairseq.py +++ b/TTS/tts/utils/fairseq.py @@ -2,7 +2,7 @@ import torch def rehash_fairseq_vits_checkpoint(checkpoint_file): - chk = torch.load(checkpoint_file, map_location=torch.device('cpu'))["model"] + chk = torch.load(checkpoint_file, map_location=torch.device("cpu"))["model"] new_chk = {} for k, v in chk.items(): if "enc_p." in k: