def setUp(self):
self.tokenizer = TTSTokenizer(use_phonemes=False,
characters=Graphemes())
self.ph = ESpeak("tr", backend="espeak")
self.tokenizer_ph = TTSTokenizer(use_phonemes=True,
characters=IPAPhonemes(),
phonemizer=self.ph)
def test_not_found_characters(self):
self.ph = ESpeak("en-us")
tokenizer_local = TTSTokenizer(use_phonemes=True,
characters=IPAPhonemes(),
phonemizer=self.ph)
self.assertEqual(len(self.tokenizer.not_found_characters), 0)
text = "Yolk of one egg beaten light"
ids = tokenizer_local.text_to_ids(text)
text_hat = tokenizer_local.ids_to_text(ids)
self.assertEqual(tokenizer_local.not_found_characters, ["̩"])
self.assertEqual(text_hat, "jˈoʊk ʌv wˈʌn ˈɛɡ bˈiːʔn lˈaɪt")
def _create_dataloader(self, batch_size, r, bgs, start_by_longest=False):
# load dataset
meta_data_train, meta_data_eval = load_tts_samples(dataset_config,
eval_split=True,
eval_split_size=0.2)
items = meta_data_train + meta_data_eval
tokenizer, _ = TTSTokenizer.init_from_config(c)
dataset = TTSDataset(
outputs_per_step=r,
compute_linear_spec=True,
return_wav=True,
tokenizer=tokenizer,
ap=self.ap,
samples=items,
batch_group_size=bgs,
min_text_len=c.min_text_len,
max_text_len=c.max_text_len,
min_audio_len=c.min_audio_len,
max_audio_len=c.max_audio_len,
start_by_longest=start_by_longest,
)
dataloader = DataLoader(
dataset,
batch_size=batch_size,
shuffle=False,
collate_fn=dataset.collate_fn,
drop_last=True,
num_workers=c.num_loader_workers,
)
return dataloader, dataset
def test_init_from_config(self):
@dataclass
class Characters(Coqpit):
characters_class: str = None
characters: str = _phonemes
punctuations: str = _punctuations
pad: str = _pad
eos: str = _eos
bos: str = _bos
blank: str = _blank
is_unique: bool = True
is_sorted: bool = True
@dataclass
class TokenizerConfig(Coqpit):
enable_eos_bos_chars: bool = True
use_phonemes: bool = True
add_blank: bool = False
characters: str = Characters()
phonemizer: str = "espeak"
phoneme_language: str = "tr"
text_cleaner: str = "phoneme_cleaners"
characters = Characters()
tokenizer_ph, _ = TTSTokenizer.init_from_config(TokenizerConfig())
tokenizer_ph.phonemizer.backend = "espeak"
text = "Bu bir Örnek."
text_ph = "<BOS>" + self.ph.phonemize(text, separator="") + "<EOS>"
ids = tokenizer_ph.text_to_ids(text)
test_hat = tokenizer_ph.ids_to_text(ids)
self.assertEqual(text_ph, test_hat)
def init_from_config(config: Coqpit):
"""Initialize model from config."""
from TTS.utils.audio import AudioProcessor
ap = AudioProcessor.init_from_config(config)
tokenizer = TTSTokenizer.init_from_config(config)
speaker_manager = SpeakerManager.init_from_config(config)
return BaseTacotron(config, ap, tokenizer, speaker_manager)
def init_from_config(config: "ForwardTTSConfig",
samples: Union[List[List], List[Dict]] = None):
"""Initiate model from config
Args:
config (ForwardTTSConfig): Model config.
samples (Union[List[List], List[Dict]]): Training samples to parse speaker ids for training.
Defaults to None.
"""
from TTS.utils.audio import AudioProcessor
ap = AudioProcessor.init_from_config(config)
tokenizer, new_config = TTSTokenizer.init_from_config(config)
speaker_manager = SpeakerManager.init_from_config(config, samples)
return ForwardTTS(new_config, ap, tokenizer, speaker_manager)
def init_from_config(config: "GlowTTSConfig",
samples: Union[List[List], List[Dict]] = None,
verbose=True):
"""Initiate model from config
Args:
config (VitsConfig): Model config.
samples (Union[List[List], List[Dict]]): Training samples to parse speaker ids for training.
Defaults to None.
verbose (bool): If True, print init messages. Defaults to True.
"""
from TTS.utils.audio import AudioProcessor
ap = AudioProcessor.init_from_config(config, verbose)
tokenizer, new_config = TTSTokenizer.init_from_config(config)
speaker_manager = SpeakerManager.init_from_config(config, samples)
return GlowTTS(new_config, ap, tokenizer, speaker_manager)
def setup_loader(ap, r, verbose=False):
tokenizer, _ = TTSTokenizer.init_from_config(c)
dataset = TTSDataset(
outputs_per_step=r,
compute_linear_spec=False,
samples=meta_data,
tokenizer=tokenizer,
ap=ap,
batch_group_size=0,
min_text_len=c.min_text_len,
max_text_len=c.max_text_len,
min_audio_len=c.min_audio_len,
max_audio_len=c.max_audio_len,
phoneme_cache_path=c.phoneme_cache_path,
precompute_num_workers=0,
use_noise_augment=False,
verbose=verbose,
speaker_id_mapping=speaker_manager.ids
if c.use_speaker_embedding else None,
d_vector_mapping=speaker_manager.embeddings
if c.use_d_vector_file else None,
)
if c.use_phonemes and c.compute_input_seq_cache:
# precompute phonemes to have a better estimate of sequence lengths.
dataset.compute_input_seq(c.num_loader_workers)
dataset.preprocess_samples()
loader = DataLoader(
dataset,
batch_size=c.batch_size,
shuffle=False,
collate_fn=dataset.collate_fn,
drop_last=False,
sampler=None,
num_workers=c.num_loader_workers,
pin_memory=False,
)
return loader
],
sort_by_audio_len=True,
max_seq_len=500000,
output_path=output_path,
datasets=[dataset_config],
)
# INITIALIZE THE AUDIO PROCESSOR
# Audio processor is used for feature extraction and audio I/O.
# It mainly serves to the dataloader and the training loggers.
ap = AudioProcessor.init_from_config(config)
# INITIALIZE THE TOKENIZER
# Tokenizer is used to convert text to sequences of token IDs.
# If characters are not defined in the config, default characters are passed to the config
tokenizer, config = TTSTokenizer.init_from_config(config)
# LOAD DATA SAMPLES
# Each sample is a list of ```[text, audio_file_path, speaker_name]```
# You can define your custom sample loader returning the list of samples.
# Or define your custom formatter and pass it to the `load_tts_samples`.
# Check `TTS.tts.datasets.load_tts_samples` for more details.
train_samples, eval_samples = load_tts_samples(
dataset_config,
eval_split=True,
eval_split_max_size=config.eval_split_max_size,
eval_split_size=config.eval_split_size,
)
# init model
model = ForwardTTS(config, ap, tokenizer)
class TestTTSTokenizer(unittest.TestCase):
def setUp(self):
self.tokenizer = TTSTokenizer(use_phonemes=False,
characters=Graphemes())
self.ph = ESpeak("tr", backend="espeak")
self.tokenizer_ph = TTSTokenizer(use_phonemes=True,
characters=IPAPhonemes(),
phonemizer=self.ph)
def test_encode_decode_graphemes(self):
text = "This is, a test."
ids = self.tokenizer.encode(text)
test_hat = self.tokenizer.decode(ids)
self.assertEqual(text, test_hat)
self.assertEqual(len(ids), len(text))
def test_text_to_ids_phonemes(self):
# TODO: note sure how to extend to cover all the languages and phonemizer.
text = "Bu bir Örnek."
text_ph = self.ph.phonemize(text, separator="")
ids = self.tokenizer_ph.text_to_ids(text)
test_hat = self.tokenizer_ph.ids_to_text(ids)
self.assertEqual(text_ph, test_hat)
def test_text_to_ids_phonemes_with_eos_bos(self):
text = "Bu bir Örnek."
self.tokenizer_ph.use_eos_bos = True
text_ph = IPAPhonemes().bos + self.ph.phonemize(
text, separator="") + IPAPhonemes().eos
ids = self.tokenizer_ph.text_to_ids(text)
test_hat = self.tokenizer_ph.ids_to_text(ids)
self.assertEqual(text_ph, test_hat)
def test_text_to_ids_phonemes_with_eos_bos_and_blank(self):
text = "Bu bir Örnek."
self.tokenizer_ph.use_eos_bos = True
self.tokenizer_ph.add_blank = True
text_ph = "<BOS><BLNK>b<BLNK>ʊ<BLNK> <BLNK>b<BLNK>ɪ<BLNK>r<BLNK> <BLNK>œ<BLNK>r<BLNK>n<BLNK>ˈ<BLNK>ɛ<BLNK>c<BLNK>.<BLNK><EOS>"
ids = self.tokenizer_ph.text_to_ids(text)
text_hat = self.tokenizer_ph.ids_to_text(ids)
self.assertEqual(text_ph, text_hat)
def test_print_logs(self):
self.tokenizer.print_logs()
self.tokenizer_ph.print_logs()
def test_not_found_characters(self):
self.ph = ESpeak("en-us")
tokenizer_local = TTSTokenizer(use_phonemes=True,
characters=IPAPhonemes(),
phonemizer=self.ph)
self.assertEqual(len(self.tokenizer.not_found_characters), 0)
text = "Yolk of one egg beaten light"
ids = tokenizer_local.text_to_ids(text)
text_hat = tokenizer_local.ids_to_text(ids)
self.assertEqual(tokenizer_local.not_found_characters, ["̩"])
self.assertEqual(text_hat, "jˈoʊk ʌv wˈʌn ˈɛɡ bˈiːʔn lˈaɪt")
def test_init_from_config(self):
@dataclass
class Characters(Coqpit):
characters_class: str = None
characters: str = _phonemes
punctuations: str = _punctuations
pad: str = _pad
eos: str = _eos
bos: str = _bos
blank: str = _blank
is_unique: bool = True
is_sorted: bool = True
@dataclass
class TokenizerConfig(Coqpit):
enable_eos_bos_chars: bool = True
use_phonemes: bool = True
add_blank: bool = False
characters: str = Characters()
phonemizer: str = "espeak"
phoneme_language: str = "tr"
text_cleaner: str = "phoneme_cleaners"
characters = Characters()
tokenizer_ph, _ = TTSTokenizer.init_from_config(TokenizerConfig())
tokenizer_ph.phonemizer.backend = "espeak"
text = "Bu bir Örnek."
text_ph = "<BOS>" + self.ph.phonemize(text, separator="") + "<EOS>"
ids = tokenizer_ph.text_to_ids(text)
test_hat = tokenizer_ph.ids_to_text(ids)
self.assertEqual(text_ph, test_hat)
