def evaluate(model, neval_batches):
model.eval()
cnt = 0
elapsed = 0
filename = config.validation_files
with open(filename, 'r') as file:
lines = file.readlines()
# Batches
for line in lines:
tokens = line.strip().split('|')
text = tokens[1]
sequence = np.array(text_to_sequence(text))[None, :]
sequence = torch.from_numpy(sequence).long()
start = time.time()
mel_outputs, mel_outputs_postnet, _, alignments = model.inference(
sequence)
end = time.time()
elapsed = elapsed + (end - start)
cnt += 1
print('.', end='')
if cnt >= neval_batches:
break
print('\nElapsed: {}{:.5f}{} sec'.format(bcolors.OKGREEN, elapsed,
bcolors.ENDC))
return
def get_text(self, text):
text = text_to_sequence(text)
text_norm = torch.IntTensor(text)
return text_norm
checkpoint = 'BEST_checkpoint.tar'
checkpoint = torch.load(checkpoint)
model = checkpoint['model']
model.eval()
waveglow_path = 'waveglow_256channels.pt'
waveglow = torch.load(waveglow_path)['model']
waveglow.cuda().eval().half()
for k in waveglow.convinv:
k.float()
denoiser = Denoiser(waveglow)
text = "相对论直接和间接的催生了量子力学的诞生 也为研究微观世界的高速运动确立了全新的数学模型"
text = pinyin.get(text, format="numerical", delimiter=" ")
print(text)
sequence = np.array(text_to_sequence(text))[None, :]
sequence = torch.autograd.Variable(
torch.from_numpy(sequence)).cuda().long()
mel_outputs, mel_outputs_postnet, _, alignments = model.inference(sequence)
plot_data((mel_outputs.float().data.cpu().numpy()[0],
mel_outputs_postnet.float().data.cpu().numpy()[0],
alignments.float().data.cpu().numpy()[0].T))
ensure_folder('images')
plt.savefig('images/mel_spec.jpg')
mel_outputs_postnet = mel_outputs_postnet.type(torch.float16)
with torch.no_grad():
audio = waveglow.infer(mel_outputs_postnet, sigma=0.666)
if __name__ == '__main__':
print('Reading vocab...')
in_vocab = read_vocab()
in_vocab += ['<unk>', '<EOS>', '<mask>']
out_vocab = ['(', ')', '<TOK>', '<EOS>']
print('Done.')
print('Reading test data...')
BATCH_SIZE = 128
_, X_test = ptb(section='wsj_23', directory='data/', column=0)
_, y_test = ptb(section='wsj_23', directory='data/', column=1)
X_test, y_test = sort_by_len(X_test, y_test)
X_test_raw, _ = batch(X_test, batch_size=BATCH_SIZE, mask='<mask>')
y_test_raw, _ = batch(y_test, batch_size=BATCH_SIZE, mask='<mask>')
X_test_seq, word_to_n, n_to_word = text_to_sequence(X_test, in_vocab)
y_test_seq, _, _ = text_to_sequence(y_test, out_vocab)
X_test_seq, X_test_masks = batch(X_test_seq, batch_size=BATCH_SIZE, mask=len(in_vocab)-1)
y_test_seq, y_test_masks = batch(y_test_seq, batch_size=BATCH_SIZE, mask=len(in_vocab)-1)
print('Done.')
print('Building model...')
collection = dy.ParameterCollection()
seq2seq = Seq2SeqAttention(collection, len(in_vocab), len(out_vocab))
print('Done.')
print('Loading model...')
RUN = 'runs/baseline'
checkpoint = os.path.join(RUN, 'baseline.model')
print('Loading from %s.' % checkpoint)
collection.populate(checkpoint)