def synthesize(speaker, input_file, output_file, params):
print("[Encoding]")
from io_modules.dataset import Dataset
from io_modules.dataset import Encodings
from models.encoder import Encoder
from trainers.encoder import Trainer
encodings = Encodings()
encodings.load('data/models/encoder.encodings')
encoder = Encoder(params, encodings, runtime=True)
encoder.load('data/models/rnn_encoder')
seq = create_lab_input(input_file, speaker)
mgc, att = encoder.generate(seq)
_render_spectrogram(mgc, output_file + '.png')
print("[Vocoding]")
from models.vocoder import Vocoder
from trainers.vocoder import Trainer
vocoder = Vocoder(params, runtime=True)
vocoder.load('data/models/rnn_vocoder')
import time
start = time.time()
signal = vocoder.synthesize(mgc,
batch_size=1000,
temperature=params.temperature)
stop = time.time()
sys.stdout.write(" execution time=" + str(stop - start))
sys.stdout.write('\n')
sys.stdout.flush()
from io_modules.dataset import DatasetIO
dio = DatasetIO()
enc = dio.b16_dec(signal, discreete=True)
dio.write_wave(output_file, enc, params.target_sample_rate)
def synth_devset(self,
batch_size,
target_sample_rate,
sample=True,
temperature=1.0):
sys.stdout.write('\tSynthesizing devset\n')
file_index = 1
for file in self.devset.files[:5]:
sys.stdout.write("\t\t" + str(file_index) + "/" +
str(len(self.devset.files)) +
" processing file " + file)
sys.stdout.flush()
file_index += 1
mgc_file = file + ".mgc.npy"
mgc = np.load(mgc_file)
import time
start = time.time()
synth = self.vocoder.synthesize(mgc,
batch_size,
sample=sample,
temperature=temperature)
stop = time.time()
sys.stdout.write(" execution time=" + str(stop - start))
sys.stdout.write('\n')
sys.stdout.flush()
dio = DatasetIO()
if self.use_ulaw:
enc = dio.ulaw_decode(synth, discreete=True)
else:
enc = dio.b16_dec(synth, discreete=True)
output_file = 'data/output/' + file[file.rfind('/') + 1:] + '.wav'
dio.write_wave(output_file, enc, target_sample_rate)