def setup_loader(ap, r, is_val=False, verbose=False, speaker_mapping=None):
if is_val and not c.run_eval:
loader = None
else:
dataset = MyDataset(
r,
c.text_cleaner,
compute_linear_spec=c.model.lower() == 'tacotron',
meta_data=meta_data_eval if is_val else meta_data_train,
ap=ap,
tp=c.characters if 'characters' in c.keys() else None,
batch_group_size=0 if is_val else c.batch_group_size *
c.batch_size,
min_seq_len=c.min_seq_len,
max_seq_len=c.max_seq_len,
phoneme_cache_path=c.phoneme_cache_path,
use_phonemes=c.use_phonemes,
phoneme_language=c.phoneme_language,
enable_eos_bos=c.enable_eos_bos_chars,
verbose=verbose,
speaker_mapping=speaker_mapping if c.use_speaker_embedding
and c.use_external_speaker_embedding_file else None)
sampler = DistributedSampler(dataset) if num_gpus > 1 else None
loader = DataLoader(
dataset,
batch_size=c.eval_batch_size if is_val else c.batch_size,
shuffle=False,
collate_fn=dataset.collate_fn,
drop_last=False,
sampler=sampler,
num_workers=c.num_val_loader_workers
if is_val else c.num_loader_workers,
pin_memory=False)
return loader
def setup_loader(ap, r, is_val=False, verbose=False, dataset=None):
if is_val and not c.run_eval:
loader = None
else:
if dataset is None:
dataset = MyDataset(
r,
c.text_cleaner,
compute_linear_spec=c.model.lower() == "tacotron",
meta_data=meta_data_eval if is_val else meta_data_train,
ap=ap,
tp=c.characters if "characters" in c.keys() else None,
add_blank=c["add_blank"] if "add_blank" in c.keys() else False,
batch_group_size=0 if is_val else c.batch_group_size *
c.batch_size,
min_seq_len=c.min_seq_len,
max_seq_len=c.max_seq_len,
phoneme_cache_path=c.phoneme_cache_path,
use_phonemes=c.use_phonemes,
phoneme_language=c.phoneme_language,
enable_eos_bos=c.enable_eos_bos_chars,
verbose=verbose,
speaker_mapping=(speaker_mapping if
(c.use_speaker_embedding
and c.use_external_speaker_embedding_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.sort_items()
sampler = DistributedSampler(dataset) if num_gpus > 1 else None
loader = DataLoader(
dataset,
batch_size=c.eval_batch_size if is_val else c.batch_size,
shuffle=False,
collate_fn=dataset.collate_fn,
drop_last=False,
sampler=sampler,
num_workers=c.num_val_loader_workers
if is_val else c.num_loader_workers,
pin_memory=False,
)
return loader
def setup_loader(ap, r, is_val=False, verbose=False):
if is_val and not config.run_eval:
loader = None
else:
dataset = MyDataset(
r,
config.text_cleaner,
compute_linear_spec=False,
meta_data=meta_data_eval if is_val else meta_data_train,
ap=ap,
tp=config.characters,
add_blank=config["add_blank"],
batch_group_size=0 if is_val else config.batch_group_size *
config.batch_size,
min_seq_len=config.min_seq_len,
max_seq_len=config.max_seq_len,
phoneme_cache_path=config.phoneme_cache_path,
use_phonemes=config.use_phonemes,
phoneme_language=config.phoneme_language,
enable_eos_bos=config.enable_eos_bos_chars,
use_noise_augment=not is_val,
verbose=verbose,
speaker_mapping=speaker_mapping if config.use_speaker_embedding
and config.use_external_speaker_embedding_file else None,
)
if config.use_phonemes and config.compute_input_seq_cache:
# precompute phonemes to have a better estimate of sequence lengths.
dataset.compute_input_seq(config.num_loader_workers)
dataset.sort_items()
sampler = DistributedSampler(dataset) if num_gpus > 1 else None
loader = DataLoader(
dataset,
batch_size=config.eval_batch_size if is_val else config.batch_size,
shuffle=False,
collate_fn=dataset.collate_fn,
drop_last=False,
sampler=sampler,
num_workers=config.num_val_loader_workers
if is_val else config.num_loader_workers,
pin_memory=False,
)
return loader
def setup_loader(ap, r, verbose=False):
dataset = MyDataset(
r,
c.text_cleaner,
compute_linear_spec=False,
meta_data=meta_data,
ap=ap,
tp=c.characters if "characters" in c.keys() else None,
add_blank=c["add_blank"] if "add_blank" in c.keys() else False,
batch_group_size=0,
min_seq_len=c.min_seq_len,
max_seq_len=c.max_seq_len,
phoneme_cache_path=c.phoneme_cache_path,
use_phonemes=c.use_phonemes,
phoneme_language=c.phoneme_language,
enable_eos_bos=c.enable_eos_bos_chars,
use_noise_augment=False,
verbose=verbose,
speaker_mapping=speaker_mapping if c.use_speaker_embedding and c.use_external_speaker_embedding_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.sort_items()
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
num_chars = len(phonemes) if C.use_phonemes else len(symbols)
# TODO: handle multi-speaker
model = setup_model(num_chars, num_speakers=0, c=C)
model, _ = load_checkpoint(model, args.model_path, None, args.use_cuda)
model.eval()
# data loader
preprocessor = importlib.import_module('TTS.tts.datasets.preprocess')
preprocessor = getattr(preprocessor, args.dataset)
meta_data = preprocessor(args.data_path, args.dataset_metafile)
dataset = MyDataset(
model.decoder.r,
C.text_cleaner,
compute_linear_spec=False,
ap=ap,
meta_data=meta_data,
tp=C.characters if 'characters' in C.keys() else None,
add_blank=C['add_blank'] if 'add_blank' in C.keys() else False,
use_phonemes=C.use_phonemes,
phoneme_cache_path=C.phoneme_cache_path,
phoneme_language=C.phoneme_language,
enable_eos_bos=C.enable_eos_bos_chars)
dataset.sort_items()
loader = DataLoader(dataset,
batch_size=args.batch_size,
num_workers=4,
collate_fn=dataset.collate_fn,
shuffle=False,
drop_last=False)
# compute attentions
