def test_arpabet():
assert to_sequence('{AE0 D}') == [70, 91, 1]
assert round_trip('{AE0 D V ER1 S}') == '{AE0 D V ER1 S}~'
assert round_trip('{AE0 D V ER1 S} circumstances') == '{AE0 D V ER1 S} circumstances~'
assert round_trip('In {AE0 D V ER1 S} circumstances') == 'In {AE0 D V ER1 S} circumstances~'
assert round_trip('{AE0 D V ER1 S} {AE0 D S}') == '{AE0 D V ER1 S} {AE0 D S}~'
assert round_trip('X {AE0 D} Y {AE0 D} Z') == 'X {AE0 D} Y {AE0 D} Z~'
def _get_next_example(self):
'''Loads a single example (input, mel_target, linear_target, cost) from disk'''
if self._offset >= len(self._metadata):
self._offset = 0
random.shuffle(self._metadata)
meta = self._metadata[self._offset]
self._offset += 1
text = meta[3]
if self._cmudict and random.random() < _p_cmudict:
text = ' '.join(
[self._maybe_get_arpabet(word) for word in text.split(' ')])
input_data = np.asarray(textinput.to_sequence(text), dtype=np.int32)
if self.cache_targets:
if (meta[0] in self._cached_spec_targets) and (
meta[1] in self._cached_mel_targets):
linear_target = self._cached_spec_targets[meta[0]]
mel_target = self._cached_mel_targets[meta[1]]
else:
linear_target = np.load(os.path.join(self._datadir, meta[0]))
mel_target = np.load(os.path.join(self._datadir, meta[1]))
if len(
self._cached_spec_targets
) < self.n_cache_audio_targets and len(
self._cached_mel_targets) < self.n_cache_audio_targets:
self._cached_spec_targets[meta[0]] = linear_target
self._cached_mel_targets[meta[1]] = mel_target
else:
linear_target = np.load(os.path.join(self._datadir, meta[0]))
mel_target = np.load(os.path.join(self._datadir, meta[1]))
return (input_data, mel_target, linear_target, len(linear_target))
def synthesize(self, text):
with chainer.using_config('train', False):
seq = textinput.to_sequence(
text,
force_lowercase=hparams.force_lowercase,
expand_abbreviations=hparams.expand_abbreviations)
spec = self.model.output(seq)
out = io.BytesIO()
audio.save_wav(audio.inv_spectrogram(spec.T), out)
return out.getvalue()
def _get_next_example(self, offset):
'''Loads a single example (input, mel_target, linear_target, cost) from disk'''
meta = self._metadata[offset]
text = meta[3]
if self._cmudict and random.random() < _p_cmudict:
text = ' '.join(
[self._maybe_get_arpabet(word) for word in text.split(' ')])
input_data = np.asarray(textinput.to_sequence(text), dtype=np.int32)
linear_target = np.load(os.path.join(self._datadir, meta[0]))
mel_target = np.load(os.path.join(self._datadir, meta[1]))
return (input_data, mel_target, linear_target, len(linear_target))
def synthesize(self, text):
seq = textinput.to_sequence(
text,
force_lowercase=hparams.force_lowercase,
expand_abbreviations=hparams.expand_abbreviations)
feed_dict = {
self.model.inputs: [np.asarray(seq, dtype=np.int32)],
self.model.input_lengths: np.asarray([len(seq)], dtype=np.int32)
}
spec, alignments = self.session.run(
[self.model.linear_outputs[0], self.model.alignments[0]],
feed_dict=feed_dict)
out = io.BytesIO()
audio.save_wav(audio.inv_spectrogram(spec.T), out)
return out.getvalue(), alignments
def test_to_sequence():
assert to_sequence('') == [1]
assert to_sequence('H', force_lowercase=False) == [9, 1]
assert to_sequence('H', force_lowercase=True) == [35, 1]
assert to_sequence('Hi.', force_lowercase=False) == [9, 36, 60, 1]
def round_trip(x): return to_string(to_sequence(x, force_lowercase=False, expand_abbreviations=True))