def _create_random_model(self):
config = load_config(
os.path.join(get_tests_output_path(), 'dummy_model_config.json'))
num_chars = len(phonemes) if config.use_phonemes else len(symbols)
model = setup_model(num_chars, 0, config)
output_path = os.path.join(get_tests_output_path())
save_checkpoint(model, None, None, None, output_path, 10, 10)
def load_tts(self, tts_checkpoint, tts_config, use_cuda):
print(" > Loading TTS model ...")
print(" | > model config: ", tts_config)
print(" | > checkpoint file: ", tts_checkpoint)
self.tts_config = load_config(tts_config)
self.use_phonemes = self.tts_config.use_phonemes
self.ap = AudioProcessor(**self.tts_config.audio)
if self.use_phonemes:
self.input_size = len(phonemes)
else:
self.input_size = len(symbols)
# TODO: fix this for multi-speaker model - load speakers
if self.config.tts_speakers is not None:
self.tts_speakers = load_speaker_mapping(self.config.tts_speakers)
num_speakers = len(self.tts_speakers)
else:
num_speakers = 0
self.tts_model = setup_model(self.input_size,
num_speakers=num_speakers,
c=self.tts_config)
# load model state
cp = torch.load(tts_checkpoint, map_location=torch.device('cpu'))
# load the model
self.tts_model.load_state_dict(cp['model'])
if use_cuda:
self.tts_model.cuda()
self.tts_model.eval()
self.tts_model.decoder.max_decoder_steps = 3000
if 'r' in cp:
self.tts_model.decoder.set_r(cp['r'])
def load_tts(self, model_path, model_file, model_config, use_cuda):
tts_config = os.path.join(model_path, model_config)
self.model_file = os.path.join(model_path, model_file)
print(" > Loading TTS model ...")
print(" | > model config: ", tts_config)
print(" | > model file: ", model_file)
self.tts_config = load_config(tts_config)
self.use_phonemes = self.tts_config.use_phonemes
self.ap = AudioProcessor(**self.tts_config.audio)
if self.use_phonemes:
self.input_size = len(phonemes)
else:
self.input_size = len(symbols)
# load speakers
if self.config.tts_speakers is not None:
self.tts_speakers = load_speaker_mapping(
os.path.join(model_path, self.config.tts_speakers))
num_speakers = len(self.tts_speakers)
else:
num_speakers = 0
self.tts_model = setup_model(self.input_size,
num_speakers=num_speakers,
c=self.tts_config)
# load model state
cp = torch.load(self.model_file)
# load the model
self.tts_model.load_state_dict(cp['model'])
if use_cuda:
self.tts_model.cuda()
self.tts_model.eval()
self.tts_model.decoder.max_decoder_steps = 3000
if 'r' in cp and self.tts_config.model in ["Tacotron", "TacotronGST"]:
self.tts_model.decoder.set_r(cp['r'])
def _create_random_model(self):
# pylint: disable=global-statement
global symbols, phonemes
config = load_config(
os.path.join(get_tests_output_path(), 'dummy_model_config.json'))
if 'characters' in config.keys():
symbols, phonemes = make_symbols(**config.characters)
num_chars = len(phonemes) if config.use_phonemes else len(symbols)
model = setup_model(num_chars, 0, config)
output_path = os.path.join(get_tests_output_path())
save_checkpoint(model, None, None, None, output_path, 10, 10)
def setup():
use_cuda = True
# model paths
TTS_MODEL = "tts_model.pth.tar"
TTS_CONFIG = "config.json"
VOCODER_MODEL = "vocoder_model.pth.tar"
VOCODER_CONFIG = "config_vocoder.json"
# Load configs
TTS_CONFIG = load_config(TTS_CONFIG)
VOCODER_CONFIG = load_config(VOCODER_CONFIG)
ap = AudioProcessor(**TTS_CONFIG.audio)
# LOAD TTS MODEL
# multi speaker
speaker_id = None
speakers = []
# load the model
num_chars = len(phonemes) if TTS_CONFIG.use_phonemes else len(symbols)
model = setup_model(num_chars, len(speakers), TTS_CONFIG)
# load model state
cp = torch.load(TTS_MODEL, map_location=torch.device('cpu'))
# load the model
model.load_state_dict(cp['model'])
if use_cuda:
model.cuda()
model.eval()
# set model stepsize
if 'r' in cp:
model.decoder.set_r(cp['r'])
from TTS.vocoder.utils.generic_utils import setup_generator
# LOAD VOCODER MODEL
vocoder_model = setup_generator(VOCODER_CONFIG)
vocoder_model.load_state_dict(torch.load(VOCODER_MODEL, map_location="cpu")["model"])
vocoder_model.remove_weight_norm()
vocoder_model.inference_padding = 0
ap_vocoder = AudioProcessor(**VOCODER_CONFIG['audio'])
if use_cuda:
vocoder_model.cuda()
vocoder_model.eval()
return model, vocoder_model, speaker_id, TTS_CONFIG, use_cuda, ap
def __init__(self, use_cuda=False, verbose=False):
self.use_cuda = use_cuda
self.verbose = verbose
# load configs
self.TTS_CONFIG = load_config(TTS_CONFIG)
self.VOCODER_CONFIG = load_config(VOCODER_CONFIG)
# load the audio processor
self.ap = AudioProcessor(**self.TTS_CONFIG.audio)
# LOAD TTS MODEL
self.speaker_id = None
speakers = []
# load the model
num_chars = len(phonemes) if self.TTS_CONFIG.use_phonemes else len(symbols)
self.model = setup_model(num_chars, len(speakers), self.TTS_CONFIG)
# load model state
cp = torch.load(TTS_MODEL, map_location=torch.device('cpu'))
# load the model
self.model.load_state_dict(cp['model'])
if self.use_cuda:
self.model.cuda()
self.model.eval()
# set model stepsize
if 'r' in cp:
self.model.decoder.set_r(cp['r'])
# LOAD VOCODER MODEL
self.vocoder_model = setup_generator(self.VOCODER_CONFIG)
self.vocoder_model.load_state_dict(torch.load(VOCODER_MODEL, map_location="cpu")["model"])
self.vocoder_model.remove_weight_norm()
self.vocoder_model.inference_padding = 0
ap_vocoder = AudioProcessor(**self.VOCODER_CONFIG['audio'])
if self.use_cuda:
self.vocoder_model.cuda()
self.vocoder_model.eval()
def __init__(self, TTS_MODEL, TTS_CONFIG, VOCODER_MODEL, VOCODER_CONFIG,
use_cuda, use_gl):
self.use_cuda = use_cuda
self.use_gl = use_gl
# model paths
self.tts_config = load_config(TTS_CONFIG)
vocoder_config = load_config(VOCODER_CONFIG)
# load audio processor
self.ap = AudioProcessor(**self.tts_config.audio)
# LOAD TTS MODEL
# multi speaker
self.speaker_id = None
speakers = []
# load the model
num_chars = len(phonemes) if self.tts_config.use_phonemes else len(
symbols)
self.model = setup_model(num_chars, len(speakers), self.tts_config)
# load model state
self.cp = torch.load(TTS_MODEL, map_location=torch.device('cpu'))
# load the model
self.model.load_state_dict(self.cp['model'])
if self.use_cuda:
self.model.cuda()
self.model.train(False)
self.model.eval()
# set model stepsize
if 'r' in self.cp:
self.model.decoder.set_r(self.cp['r'])
# LOAD VOCODER MODEL
self.vocoder_model = setup_generator(vocoder_config)
self.vocoder_model.load_state_dict(
torch.load(VOCODER_MODEL, map_location="cpu")["model"])
self.vocoder_model.remove_weight_norm()
self.vocoder_model.inference_padding = 0
#ap_vocoder = AudioProcessor(**vocoder_config['audio'])
if use_cuda:
self.vocoder_model.cuda()
self.vocoder_model.train(False)
self.vocoder_model.eval()
#get sample rate
self.sample_rate = self.ap.sample_rate
gc.collect(2)
def _load_tts(self):
# LOAD TTS MODEL
from TTS.utils.text.symbols import symbols, phonemes
from TTS.utils.audio import AudioProcessor
from TTS.utils.generic_utils import setup_model
# load the model
num_chars = len(phonemes) if self.tts_config.use_phonemes else len(symbols)
self.tts_model = setup_model(num_chars, self.tts_config)
# load the audio processor
self._ap = AudioProcessor(**self.tts_config.audio)
# load model state
cp = torch.load(self.tts_model_path, map_location=lambda storage, loc: storage)
# load the model
self.tts_model.load_state_dict(cp['model'])
self.tts_model.to(self.device)
self.tts_model.eval()
self.tts_model.decoder.max_decoder_steps = 2000
def load_tts(self, tts_checkpoint, tts_config, use_cuda):
global symbols, phonemes
print(" > Loading TTS model ...")
print(" | > model config: ", tts_config)
print(" | > checkpoint file: ", tts_checkpoint)
self.tts_config = load_config(tts_config)
if 'text' in self.tts_config.keys():
symbols, phonemes = make_symbols(**self.tts_config.text)
self.use_phonemes = self.tts_config.use_phonemes
self.ap = AudioProcessor(**self.tts_config.audio)
if self.use_phonemes:
self.input_size = len(phonemes)
else:
self.input_size = len(symbols)
# load speakers
if self.config.tts_speakers is not None:
self.tts_speakers = load_speaker_mapping(
os.path.join(model_path, self.config.tts_speakers))
num_speakers = len(self.tts_speakers)
else:
num_speakers = 0
self.tts_model = setup_model(self.input_size,
num_speakers=num_speakers,
c=self.tts_config)
# load model state
cp = torch.load(tts_checkpoint, map_location=torch.device('cpu'))
# load the model
self.tts_model.load_state_dict(cp['model'])
if use_cuda:
self.tts_model.cuda()
self.tts_model.eval()
self.tts_model.decoder.max_decoder_steps = 3000
if 'r' in cp:
self.tts_model.decoder.set_r(cp['r'])
ap = AudioProcessor(**C.audio)
# if the vocabulary was passed, replace the default
if 'characters' in C.keys():
symbols, phonemes = make_symbols(**C.characters)
# load speakers
if args.speakers_json != '':
speakers = json.load(open(args.speakers_json, 'r'))
num_speakers = len(speakers)
else:
num_speakers = 0
# load the model
num_chars = len(phonemes) if C.use_phonemes else len(symbols)
model = setup_model(num_chars, num_speakers, C)
cp = torch.load(args.model_path)
model.load_state_dict(cp['model'])
model.eval()
if args.use_cuda:
model.cuda()
model.decoder.set_r(cp['r'])
# load vocoder model
if args.vocoder_path != "":
VC = load_config(args.vocoder_config_path)
ap_vocoder = AudioProcessor(**VC.audio)
bits = 10
vocoder_model = VocoderModel(rnn_dims=512,
fc_dims=512,
mode=VC.mode,
def main(args): #pylint: disable=redefined-outer-name
# Audio processor
ap = AudioProcessor(**c.audio)
# DISTRUBUTED
if num_gpus > 1:
init_distributed(args.rank, num_gpus, args.group_id,
c.distributed["backend"], c.distributed["url"])
num_chars = len(phonemes) if c.use_phonemes else len(symbols)
if c.use_speaker_embedding:
speakers = get_speakers(c.data_path, c.meta_file_train, c.dataset)
if args.restore_path:
prev_out_path = os.path.dirname(args.restore_path)
speaker_mapping = load_speaker_mapping(prev_out_path)
assert all([speaker in speaker_mapping
for speaker in speakers]), "As of now you, you cannot " \
"introduce new speakers to " \
"a previously trained model."
else:
speaker_mapping = {name: i for i, name in enumerate(speakers)}
save_speaker_mapping(OUT_PATH, speaker_mapping)
num_speakers = len(speaker_mapping)
print("Training with {} speakers: {}".format(num_speakers,
", ".join(speakers)))
else:
num_speakers = 0
model = setup_model(num_chars, num_speakers, c)
print(" | > Num output units : {}".format(ap.num_freq), flush=True)
optimizer = RAdam(model.parameters(), lr=c.lr, weight_decay=0)
if c.stopnet and c.separate_stopnet:
optimizer_st = RAdam(model.decoder.stopnet.parameters(),
lr=c.lr,
weight_decay=0)
else:
optimizer_st = None
if c.loss_masking:
criterion = L1LossMasked() if c.model in ["Tacotron", "TacotronGST"
] else MSELossMasked()
else:
criterion = nn.L1Loss() if c.model in ["Tacotron", "TacotronGST"
] else nn.MSELoss()
criterion_st = nn.BCEWithLogitsLoss() if c.stopnet else None
if args.restore_path:
checkpoint = torch.load(args.restore_path)
try:
# TODO: fix optimizer init, model.cuda() needs to be called before
# optimizer restore
# optimizer.load_state_dict(checkpoint['optimizer'])
if c.reinit_layers:
raise RuntimeError
model.load_state_dict(checkpoint['model'])
except:
print(" > Partial model initialization.")
model_dict = model.state_dict()
model_dict = set_init_dict(model_dict, checkpoint, c)
model.load_state_dict(model_dict)
del model_dict
for group in optimizer.param_groups:
group['lr'] = c.lr
print(" > Model restored from step %d" % checkpoint['step'],
flush=True)
args.restore_step = checkpoint['step']
else:
args.restore_step = 0
if use_cuda:
model = model.cuda()
criterion.cuda()
if criterion_st:
criterion_st.cuda()
# DISTRUBUTED
if num_gpus > 1:
model = apply_gradient_allreduce(model)
if c.lr_decay:
scheduler = NoamLR(optimizer,
warmup_steps=c.warmup_steps,
last_epoch=args.restore_step - 1)
else:
scheduler = None
num_params = count_parameters(model)
print("\n > Model has {} parameters".format(num_params), flush=True)
if 'best_loss' not in locals():
best_loss = float('inf')
global_step = args.restore_step
for epoch in range(0, c.epochs):
# set gradual training
if c.gradual_training is not None:
r, c.batch_size = gradual_training_scheduler(global_step, c)
c.r = r
model.decoder.set_r(r)
print(" > Number of outputs per iteration:", model.decoder.r)
train_loss, global_step = train(model, criterion, criterion_st,
optimizer, optimizer_st, scheduler, ap,
global_step, epoch)
val_loss = evaluate(model, criterion, criterion_st, ap, global_step,
epoch)
print(" | > Training Loss: {:.5f} Validation Loss: {:.5f}".format(
train_loss, val_loss),
flush=True)
target_loss = train_loss
if c.run_eval:
target_loss = val_loss
best_loss = save_best_model(model, optimizer, target_loss, best_loss,
OUT_PATH, global_step, epoch)
)
print(" > Run-time: {}".format(time.time() - t_1))
return alignment, mel_postnet_spec, stop_tokens, waveform
use_cuda = True
batched_wavernn = True
# initialize TTS
CONFIG = load_config(tts_pretrained_model_config)
print(CONFIG)
# load the model
num_chars = len(phonemes) if CONFIG.use_phonemes else len(symbols)
model = setup_model(num_chars, CONFIG)
# load the audio processor
ap = AudioProcessor(**CONFIG.audio)
# load model state
if use_cuda:
cp = torch.load(tts_pretrained_model)
else:
cp = torch.load(tts_pretrained_model, map_location=lambda storage, loc: storage)
# load the model
model.load_state_dict(cp["model"])
if use_cuda:
model.cuda()
model.eval()
print(cp["step"])
model.decoder.max_decoder_steps = 2000
def main(args): # pylint: disable=redefined-outer-name
# pylint: disable=global-variable-undefined
global meta_data_train, meta_data_eval, symbols, phonemes
# Audio processor
ap = AudioProcessor(**c.audio)
if 'characters' in c.keys():
symbols, phonemes = make_symbols(**c.characters)
# DISTRUBUTED
if num_gpus > 1:
init_distributed(args.rank, num_gpus, args.group_id,
c.distributed["backend"], c.distributed["url"])
num_chars = len(phonemes) if c.use_phonemes else len(symbols)
# load data instances
meta_data_train, meta_data_eval = load_meta_data(c.datasets)
# parse speakers
if c.use_speaker_embedding:
speakers = get_speakers(meta_data_train)
if args.restore_path:
prev_out_path = os.path.dirname(args.restore_path)
speaker_mapping = load_speaker_mapping(prev_out_path)
assert all([speaker in speaker_mapping
for speaker in speakers]), "As of now you, you cannot " \
"introduce new speakers to " \
"a previously trained model."
else:
speaker_mapping = {name: i for i, name in enumerate(speakers)}
save_speaker_mapping(OUT_PATH, speaker_mapping)
num_speakers = len(speaker_mapping)
print("Training with {} speakers: {}".format(num_speakers,
", ".join(speakers)))
else:
num_speakers = 0
model = setup_model(num_chars, num_speakers, c)
print(" | > Num output units : {}".format(ap.num_freq), flush=True)
params = set_weight_decay(model, c.wd)
optimizer = RAdam(params, lr=c.lr, weight_decay=0)
if c.stopnet and c.separate_stopnet:
optimizer_st = RAdam(model.decoder.stopnet.parameters(),
lr=c.lr,
weight_decay=0)
else:
optimizer_st = None
# setup criterion
criterion = TacotronLoss(c, stopnet_pos_weight=10.0, ga_sigma=0.4)
if args.restore_path:
checkpoint = torch.load(args.restore_path, map_location='cpu')
try:
# TODO: fix optimizer init, model.cuda() needs to be called before
# optimizer restore
# optimizer.load_state_dict(checkpoint['optimizer'])
if c.reinit_layers:
raise RuntimeError
model.load_state_dict(checkpoint['model'])
except:
print(" > Partial model initialization.")
model_dict = model.state_dict()
model_dict = set_init_dict(model_dict, checkpoint, c)
model.load_state_dict(model_dict)
del model_dict
for group in optimizer.param_groups:
group['lr'] = c.lr
print(" > Model restored from step %d" % checkpoint['step'],
flush=True)
args.restore_step = checkpoint['step']
else:
args.restore_step = 0
if use_cuda:
model.cuda()
criterion.cuda()
# DISTRUBUTED
if num_gpus > 1:
model = apply_gradient_allreduce(model)
if c.noam_schedule:
scheduler = NoamLR(optimizer,
warmup_steps=c.warmup_steps,
last_epoch=args.restore_step - 1)
else:
scheduler = None
num_params = count_parameters(model)
print("\n > Model has {} parameters".format(num_params), flush=True)
if 'best_loss' not in locals():
best_loss = float('inf')
global_step = args.restore_step
for epoch in range(0, c.epochs):
c_logger.print_epoch_start(epoch, c.epochs)
# set gradual training
if c.gradual_training is not None:
r, c.batch_size = gradual_training_scheduler(global_step, c)
c.r = r
model.decoder.set_r(r)
if c.bidirectional_decoder:
model.decoder_backward.set_r(r)
print("\n > Number of output frames:", model.decoder.r)
train_avg_loss_dict, global_step = train(model, criterion, optimizer,
optimizer_st, scheduler, ap,
global_step, epoch)
eval_avg_loss_dict = evaluate(model, criterion, ap, global_step, epoch)
c_logger.print_epoch_end(epoch, eval_avg_loss_dict)
target_loss = train_avg_loss_dict['avg_postnet_loss']
if c.run_eval:
target_loss = eval_avg_loss_dict['avg_postnet_loss']
best_loss = save_best_model(target_loss, best_loss, model, optimizer,
global_step, epoch, c.r, OUT_PATH)
def main(**kwargs):
global symbols, phonemes # pylint: disable=global-statement
current_date = date.today()
current_date = current_date.strftime("%B %d %Y")
start_time = time.time()
# read passed variables from gui
text = kwargs['text'] # text to generate speech from
use_cuda = kwargs['use_cuda'] # if gpu exists default is true
project = kwargs['project'] # path to project folder
vocoder_type = kwargs['vocoder'] # vocoder type, default is GL
use_gst = kwargs['use_gst'] # use style_wave for prosody
style_dict = kwargs['style_input'] # use style_wave for prosody
speaker_id = kwargs['speaker_id'] # name of the selected speaker
sentence_file = kwargs['sentence_file'] # path to file if generate from file
out_path = kwargs['out_path'] # path to save the output wav
batched_vocoder = True
# load speakers
speakers_file_path = Path(project, "speakers.json")
if speakers_file_path.is_file():
speaker_data = json.load(open(speakers_file_path, 'r'))
num_speakers = len(speaker_data)
#get the speaker id for selected speaker
if speaker_id >= num_speakers:
print('Speaker ID outside of number of speakers range. Using default 0.')
speaker_id = 0
speaker_name = [speaker for speaker, id in speaker_data.items() if speaker_id == id][0]
else:
speaker_name = [speaker for speaker, id in speaker_data.items() if speaker_id == id][0]
else:
speaker_name = 'Default'
num_speakers = 0
speaker_id = None
# load the config
config_path = Path(project, "config.json")
C = load_config(config_path)
if use_gst:
if style_dict is not None:
style_input = style_dict
else:
style_input = None
# load the audio processor
ap = AudioProcessor(**C.audio)
# if the vocabulary was passed, replace the default
if 'characters' in C.keys():
symbols, phonemes = make_symbols(**C.characters)
# find the tts model file in project folder
try:
tts_model_file = glob(str(Path(project, '*.pth.tar')))
if not tts_model_file:
raise FileNotFoundError
model_path = tts_model_file[0]
except FileNotFoundError:
print('[!] TTS Model not found in path: "{}"'.format(project))
# load the model
num_chars = len(phonemes) if C.use_phonemes else len(symbols)
model = setup_model(num_chars, num_speakers, C)
# if gpu is not available use cpu
model, state = load_checkpoint(model, model_path, use_cuda=use_cuda)
model.decoder.max_decoder_steps = 2000
model.eval()
print(' > Model step:', state['step'])
print(' > Model r: ', state['r'])
# load vocoder
if vocoder_type is 'MelGAN':
try:
model_file = glob(str(Path(project, 'vocoder/*.pth.tar')))
vocoder, ap_vocoder = load_vocoder(str(Path('TTS')),
str(model_file[0]),
str(Path(project, 'vocoder/config.json')),
use_cuda)
except Exception:
print('[!] Error loading vocoder: "{}"'.format(project))
sys.exit(0)
elif vocoder_type is 'WaveRNN':
try:
model_file = glob(str(Path(project, 'vocoder/*.pkl')))
vocoder, ap_vocoder = load_vocoder(str(Path('TTS')), str(model_file[0]), str(Path(project, 'config.yml')), use_cuda)
except Exception:
print('[!] Error loading vocoder: "{}"'.format(project))
sys.exit(0)
else:
vocoder, ap_vocoder = None, None
print(" > Vocoder: {}".format(vocoder_type))
print(' > Using style input: {}\n'.format(style_input))
if sentence_file != '':
with open(sentence_file, "r", encoding='utf8') as f:
list_of_sentences = [s.strip() for s in f.readlines()]
else:
list_of_sentences = [text.strip()]
# iterate over every passed sentence and synthesize
for _, tts_sentence in enumerate(list_of_sentences):
wav_list = []
# remove character which are not alphanumerical or contain ',. '
tts_sentence = clean_sentence(tts_sentence)
print(" > Text: {}".format(tts_sentence))
# build filename
current_time = datetime.now().strftime("%H%M%S")
file_name = ' '.join(tts_sentence.split(" ")[:10])
# if multiple sentences in one line -> split them
tts_sentence = split_into_sentences(tts_sentence)
# if sentence was split in sub-sentences -> iterate over them
for sentence in tts_sentence:
# synthesize voice
_, _, _, wav = tts(model,
vocoder,
C,
None,
sentence,
ap,
ap_vocoder,
use_cuda,
batched_vocoder,
speaker_id=speaker_id,
style_input=style_input,
figures=False)
# join sub-sentences back together and add a filler between them
wav_list += list(wav)
wav_list += [0] * 10000
wav = np.array(wav_list)
# finalize filename
file_name = "_".join([str(current_time), file_name])
file_name = file_name.translate(
str.maketrans('', '', string.punctuation.replace('_', ''))) + '.wav'
if out_path == "":
out_dir = str(Path(project, 'output', current_date, speaker_name))
out_path = os.path.join(out_dir, file_name)
else:
out_dir = os.path.dirname(out_path)
# create output directory if it doesn't exist
if not os.path.isdir(out_dir):
os.makedirs(out_dir, exist_ok=True)
# save generated wav to disk
ap.save_wav(wav, out_path)
end_time = time.time()
print(" > Run-time: {}".format(end_time - start_time))
print(" > Saving output to {}\n".format(out_path))
# load configs
TTS_CONFIG = load_config(TTS_CONFIG)
VOCODER_CONFIG = load_config(VOCODER_CONFIG)
# load the audio processor
ap = AudioProcessor(**TTS_CONFIG.audio)
# LOAD TTS MODEL
# multi speaker
speaker_id = None
speakers = []
# load the model
num_chars = len(phonemes) if TTS_CONFIG.use_phonemes else len(symbols)
model = setup_model(num_chars, len(speakers), TTS_CONFIG)
# load model state
cp = torch.load(TTS_MODEL, map_location=torch.device("cpu"))
# load the model
model.load_state_dict(cp["model"])
if use_cuda:
model.cuda()
model.eval()
# set model stepsize
if "r" in cp:
model.decoder.set_r(cp["r"])