def decode(self, ids, strip_extraneous=False):
"""Transform a sequence of event indices into a performance MIDI file.
Args:
ids: List of performance event indices.
strip_extraneous: Whether to strip EOS and padding from the end of `ids`.
Returns:
Path to the temporary file where the MIDI was saved.
"""
if strip_extraneous:
ids = text_encoder.strip_ids(ids, list(range(self.num_reserved_ids)))
performance = magenta.music.Performance(
quantized_sequence=None,
steps_per_second=self._steps_per_second,
num_velocity_bins=self._num_velocity_bins)
for i in ids:
performance.append(self._encoding.decode_event(i - self.num_reserved_ids))
ns = performance.to_sequence()
_, tmp_file_path = tempfile.mkstemp('_decode.mid')
magenta.music.sequence_proto_to_midi_file(ns, tmp_file_path)
return tmp_file_path
def decode_to_note_sequence(self, ids, strip_extraneous=False):
"""Transform a sequence of event indices into a performance NoteSequence.
Args:
ids: List of performance event indices.
strip_extraneous: Whether to strip EOS and padding from the end of `ids`.
Returns:
A NoteSequence.
"""
if strip_extraneous:
ids = text_encoder.strip_ids(ids,
list(range(self.num_reserved_ids)))
# Decode indices corresponding to event n-grams back into the n-grams.
event_ids = []
for i in ids:
if i >= self.unigram_vocab_size:
event_ids += self._ngrams[i - self.unigram_vocab_size]
else:
event_ids.append(i)
performance = note_seq.Performance(
quantized_sequence=None,
steps_per_second=self._steps_per_second,
num_velocity_bins=self._num_velocity_bins)
for i in event_ids:
performance.append(
self._encoding.decode_event(i - self.num_reserved_ids))
ns = performance.to_sequence()
return ns
def decode(self, ids, strip_extraneous=False):
"""Transform a sequence of event indices into a performance MIDI file.
Args:
ids: List of performance event indices.
strip_extraneous: Whether to strip EOS and padding from the end of `ids`.
Returns:
Path to the temporary file where the MIDI was saved.
"""
if strip_extraneous:
ids = text_encoder.strip_ids(ids, list(range(self.num_reserved_ids)))
# Decode indices corresponding to event n-grams back into the n-grams.
event_ids = []
for i in ids:
if i >= self.unigram_vocab_size:
event_ids += self._ngrams[i - self.unigram_vocab_size]
else:
event_ids.append(i)
performance = magenta.music.Performance(
quantized_sequence=None,
steps_per_second=self._steps_per_second,
num_velocity_bins=self._num_velocity_bins)
for i in event_ids:
performance.append(self._encoding.decode_event(i - self.num_reserved_ids))
ns = performance.to_sequence()
_, tmp_file_path = tempfile.mkstemp('_decode.mid')
magenta.music.sequence_proto_to_midi_file(ns, tmp_file_path)
return tmp_file_path
def decode(self, ids, strip_extraneous=False):
"""Transform a sequence of event indices into a performance MIDI file.
Args:
ids: List of performance event indices.
strip_extraneous: Whether to strip EOS and padding from the end of `ids`.
Returns:
Path to the temporary file where the MIDI was saved.
"""
if strip_extraneous:
ids = text_encoder.strip_ids(ids, list(range(self.num_reserved_ids)))
# Decode indices corresponding to event n-grams back into the n-grams.
event_ids = []
for i in ids:
if i >= self.unigram_vocab_size:
event_ids += self._ngrams[i - self.unigram_vocab_size]
else:
event_ids.append(i)
performance = note_seq.Performance(
quantized_sequence=None,
steps_per_second=self._steps_per_second,
num_velocity_bins=self._num_velocity_bins)
for i in event_ids:
performance.append(self._encoding.decode_event(i - self.num_reserved_ids))
ns = performance.to_sequence()
_, tmp_file_path = tempfile.mkstemp('_decode.mid')
note_seq.sequence_proto_to_midi_file(ns, tmp_file_path)
return tmp_file_path
def decode(self, ids, strip_extraneous=False):
"""Converts a sequence of subtoken ids to a native string.
Args:
ids: a list of integers in the range [0, vocab_size)
strip_extraneous: bool, whether to strip off extraneous tokens
(EOS and PAD).
Returns:
a native string
"""
if strip_extraneous:
ids = strip_ids(ids, list(range(self._num_reserved_ids or 0)))
if isinstance(ids, list):
return self.sp.DecodeIds(list(map(lambda x: int(x), ids)))
return self.sp.DecodeIds(ids.tolist())
def decode(self, ids, strip_extraneous=False):
if strip_extraneous:
ids = strip_ids(ids, list(range(self._num_reserved_ids or 0)))
substrings = [self.all_symbols[idx] for idx in ids]
concat = "".join(substrings)
split = concat.split(EOW)
tokens = []
for subtoken in split:
if subtoken:
unescaped_token = remove_meta_symbols(subtoken)
if unescaped_token:
tokens.append(unescaped_token)
detokenized = t2t_tokenizer.decode(tokens)
return detokenized