def process(self, input_example):
tf.logging.info('Splitting %s',
input_example.features.feature['id'].bytes_list.value[0])
wav_data = input_example.features.feature['audio'].bytes_list.value[0]
ns = music_pb2.NoteSequence.FromString(
input_example.features.feature['sequence'].bytes_list.value[0])
Metrics.counter('split_wav', 'read_midi_wav_to_split').inc()
if self._split == 'test':
# For the 'test' split, use the full length audio and midi.
split_examples = split_audio_and_label_data.process_record(
wav_data,
ns,
ns.id,
min_length=0,
max_length=-1,
sample_rate=self._sample_rate)
for example in split_examples:
Metrics.counter('split_wav', 'full_example').inc()
yield example
else:
split_examples = split_audio_and_label_data.process_record(
wav_data, ns, ns.id, self._min_length, self._max_length,
self._sample_rate)
for example in split_examples:
Metrics.counter('split_wav', 'split_example').inc()
yield example
def process(self, input_example):
tf.logging.info('Splitting %s',
input_example.features.feature['id'].bytes_list.value[0])
wav_data = input_example.features.feature['audio'].bytes_list.value[0]
ns = music_pb2.NoteSequence.FromString(
input_example.features.feature['sequence'].bytes_list.value[0])
Metrics.counter('split_wav', 'read_midi_wav_to_split').inc()
if self._split == 'test':
# For the 'test' split, use the full length audio and midi.
split_examples = split_audio_and_label_data.process_record(
wav_data,
ns,
ns.id,
min_length=0,
max_length=-1,
sample_rate=self._sample_rate)
for example in split_examples:
Metrics.counter('split_wav', 'full_example').inc()
yield example
else:
split_examples = split_audio_and_label_data.process_record(
wav_data, ns, ns.id, self._min_length, self._max_length,
self._sample_rate)
for example in split_examples:
Metrics.counter('split_wav', 'split_example').inc()
yield example
def split_wav(input_example, min_length, max_length, sample_rate,
output_directory, process_for_training, load_audio_with_librosa):
"""Splits wav and midi files for the dataset."""
tf.logging.info('Splitting %s',
input_example.features.feature['id'].bytes_list.value[0])
wav_data = input_example.features.feature['audio'].bytes_list.value[0]
ns = music_pb2.NoteSequence.FromString(
input_example.features.feature['sequence'].bytes_list.value[0])
Metrics.counter('split_wav', 'read_midi_wav_to_split').inc()
if not process_for_training:
split_examples = split_audio_and_label_data.process_record(
wav_data,
ns,
ns.id,
min_length=0,
max_length=-1,
sample_rate=sample_rate,
load_audio_with_librosa=load_audio_with_librosa)
for example in split_examples:
Metrics.counter('split_wav', 'full_example').inc()
yield example
else:
try:
split_examples = split_audio_and_label_data.process_record(
wav_data,
ns,
ns.id,
min_length=min_length,
max_length=max_length,
sample_rate=sample_rate,
load_audio_with_librosa=load_audio_with_librosa)
for example in split_examples:
Metrics.counter('split_wav', 'split_example').inc()
yield example
except AssertionError:
output_file = 'badexample-' + hashlib.md5(
ns.id).hexdigest() + '.proto'
output_path = os.path.join(output_directory, output_file)
tf.logging.error('Exception processing %s. Writing file to %s',
ns.id, output_path)
with tf.gfile.Open(output_path, 'w') as f:
f.write(input_example.SerializeToString())
raise
def generate_train_set(exclude_ids):
"""Generate the train TFRecord."""
train_file_pairs = []
for directory in train_dirs:
path = os.path.join(FLAGS.input_dir, directory)
path = os.path.join(path, '*.wav')
wav_files = glob.glob(path)
# find matching mid files
for wav_file in wav_files:
base_name_root, _ = os.path.splitext(wav_file)
mid_file = base_name_root + '.mid'
if filename_to_id(wav_file) not in exclude_ids:
train_file_pairs.append((wav_file, mid_file))
train_output_name = os.path.join(FLAGS.output_dir,
'maps_config2_train.tfrecord')
with tf.python_io.TFRecordWriter(train_output_name) as writer:
for idx, pair in enumerate(train_file_pairs):
print('{} of {}: {}'.format(idx, len(train_file_pairs), pair[0]))
# load the wav data
wav_data = tf.gfile.Open(pair[0], 'rb').read()
# load the midi data and convert to a notesequence
ns = midi_io.midi_file_to_note_sequence(pair[1])
for example in split_audio_and_label_data.process_record(
wav_data, ns, pair[0], FLAGS.min_length, FLAGS.max_length,
FLAGS.sample_rate):
writer.write(example.SerializeToString())
def generate_train_set(exclude_ids):
"""Generate the train TFRecord."""
train_file_pairs = []
for directory in train_dirs:
path = os.path.join(FLAGS.input_dir, directory)
path = os.path.join(path, '*.wav')
wav_files = glob.glob(path)
# find matching mid files
for wav_file in wav_files:
base_name_root, _ = os.path.splitext(wav_file)
mid_file = base_name_root + '.mid'
if filename_to_id(wav_file) not in exclude_ids:
train_file_pairs.append((wav_file, mid_file))
train_output_name = os.path.join(FLAGS.output_dir,
'maps_config2_train.tfrecord')
with tf.python_io.TFRecordWriter(train_output_name) as writer:
for idx, pair in enumerate(train_file_pairs):
print('{} of {}: {}'.format(idx, len(train_file_pairs), pair[0]))
# load the wav data
wav_data = tf.gfile.Open(pair[0], 'rb').read()
# load the midi data and convert to a notesequence
ns = midi_io.midi_file_to_note_sequence(pair[1])
for example in split_audio_and_label_data.process_record(
wav_data, ns, pair[0], FLAGS.min_length, FLAGS.max_length,
FLAGS.sample_rate):
writer.write(example.SerializeToString())
def process(self, input_example):
tf.logging.info('Splitting %s',
input_example.features.feature['id'].bytes_list.value[0])
wav_data = input_example.features.feature['audio'].bytes_list.value[0]
ns = music_pb2.NoteSequence.FromString(
input_example.features.feature['sequence'].bytes_list.value[0])
Metrics.counter('split_wav', 'read_midi_wav_to_split').inc()
if self._split == 'test' or self._split == 'validation':
# For the 'test' and 'validation' splits, use the full length audio and
# midi.
split_examples = split_audio_and_label_data.process_record(
wav_data,
ns,
ns.id,
min_length=0,
max_length=-1,
sample_rate=self._sample_rate)
for example in split_examples:
Metrics.counter('split_wav', 'full_example').inc()
yield example
else:
try:
split_examples = split_audio_and_label_data.process_record(
wav_data, ns, ns.id, self._min_length, self._max_length,
self._sample_rate)
for example in split_examples:
Metrics.counter('split_wav', 'split_example').inc()
yield example
except AssertionError:
output_file = 'badexample-' + hashlib.md5(ns.id).hexdigest() + '.proto'
output_path = os.path.join(self._output_directory, output_file)
tf.logging.error('Exception processing %s. Writing file to %s',
ns.id, output_path)
with tf.gfile.Open(output_path, 'w') as f:
f.write(input_example.SerializeToString())
raise
def process(self, input_example):
tf.logging.info('Splitting %s',
input_example.features.feature['id'].bytes_list.value[0])
wav_data = input_example.features.feature['audio'].bytes_list.value[0]
ns = music_pb2.NoteSequence.FromString(
input_example.features.feature['sequence'].bytes_list.value[0])
Metrics.counter('split_wav', 'read_midi_wav_to_split').inc()
if not self._chunk_files:
split_examples = split_audio_and_label_data.process_record(
wav_data,
ns,
ns.id,
min_length=0,
max_length=-1,
sample_rate=self._sample_rate)
for example in split_examples:
Metrics.counter('split_wav', 'full_example').inc()
yield example
else:
try:
split_examples = split_audio_and_label_data.process_record(
wav_data, ns, ns.id, self._min_length, self._max_length,
self._sample_rate)
for example in split_examples:
Metrics.counter('split_wav', 'split_example').inc()
yield example
except AssertionError:
output_file = 'badexample-' + hashlib.md5(ns.id).hexdigest() + '.proto'
output_path = os.path.join(self._output_directory, output_file)
tf.logging.error('Exception processing %s. Writing file to %s',
ns.id, output_path)
with tf.gfile.Open(output_path, 'w') as f:
f.write(input_example.SerializeToString())
raise
def create_example(filename):
"""Processes an audio file into an Example proto."""
wav_data = tf.gfile.Open(filename, 'rb').read()
example_list = list(
split_audio_and_label_data.process_record(
wav_data=wav_data,
ns=music_pb2.NoteSequence(),
example_id=filename,
min_length=0,
max_length=-1,
allow_empty_notesequence=True))
assert len(example_list) == 1
return example_list[0].SerializeToString()
def create_example(filename):
"""Processes an audio file into an Example proto."""
wav_data = tf.gfile.Open(filename, 'rb').read()
example_list = list(
split_audio_and_label_data.process_record(
wav_data=wav_data,
ns=music_pb2.NoteSequence(),
# decode to handle filenames with extended characters.
example_id=filename.decode('utf-8'),
min_length=0,
max_length=-1,
allow_empty_notesequence=True))
assert len(example_list) == 1
return example_list[0].SerializeToString()
def testSplitAudioLabelData(self):
wav_data, sequence = self._CreateSyntheticExample()
records = split_audio_and_label_data.process_record(
wav_data, sequence, 'test', sample_rate=SAMPLE_RATE)
for record in records:
audio = record.features.feature['audio'].bytes_list.value[0]
velocity_range = music_pb2.VelocityRange.FromString(
record.features.feature['velocity_range'].bytes_list.value[0])
note_sequence = music_pb2.NoteSequence.FromString(
record.features.feature['sequence'].bytes_list.value[0])
self.assertEqual(
np.all(
audio_io.wav_data_to_samples(audio, sample_rate=SAMPLE_RATE) ==
np.zeros(2 * SAMPLE_RATE)), True)
self.assertEqual(velocity_range.min, 20)
self.assertEqual(velocity_range.max, 80)
self.assertEqual(note_sequence.notes[0].velocity, 20)
self.assertEqual(note_sequence.notes[0].end_time, 5.)
self.assertEqual(note_sequence.notes[1].velocity, 80)
self.assertEqual(note_sequence.notes[1].end_time, 10.)
def testSplitAudioLabelData(self):
wav_data, sequence = self._CreateSyntheticExample()
records = split_audio_and_label_data.process_record(
wav_data, sequence, 'test', sample_rate=SAMPLE_RATE)
for record in records:
audio = record.features.feature['audio'].bytes_list.value[0]
velocity_range = music_pb2.VelocityRange.FromString(
record.features.feature['velocity_range'].bytes_list.value[0])
note_sequence = music_pb2.NoteSequence.FromString(
record.features.feature['sequence'].bytes_list.value[0])
self.assertEqual(
np.all(
audio_io.wav_data_to_samples(audio,
sample_rate=SAMPLE_RATE) ==
np.zeros(2 * SAMPLE_RATE)), True)
self.assertEqual(velocity_range.min, 20)
self.assertEqual(velocity_range.max, 80)
self.assertEqual(note_sequence.notes[0].velocity, 20)
self.assertEqual(note_sequence.notes[0].end_time, 5.)
self.assertEqual(note_sequence.notes[1].velocity, 80)
self.assertEqual(note_sequence.notes[1].end_time, 10.)
