Created
May 23, 2019 01:33
-
-
Save chaonan99/20686fb6e94af757ac5a4c28fa0ff018 to your computer and use it in GitHub Desktop.
Experiment with tf record file provided by onsets and frames.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import os | |
import tensorflow as tf | |
from magenta.models.onsets_frames_transcription import configs | |
from magenta.models.onsets_frames_transcription import constants | |
from magenta.models.onsets_frames_transcription import data | |
from magenta.models.onsets_frames_transcription import split_audio_and_label_data | |
from magenta.models.onsets_frames_transcription import train_util | |
from magenta.music import midi_io | |
from magenta.protobuf import music_pb2 | |
from magenta.music import sequences_lib | |
from magenta.common import flatten_maybe_padded_sequences | |
def main(): | |
config = configs.CONFIG_MAP['onsets_frames'] | |
hparams = config.hparams | |
hparams.use_cudnn = False | |
hparams.batch_size = 1 | |
examples = os.path.join(os.environ['DATA_DIR'], 'onset_maestro_v1/train.tfrecord*') | |
filenames = tf.data.Dataset.list_files(examples, shuffle=False) | |
input_dataset = tf.data.TFRecordDataset(filenames) | |
iterator = input_dataset.make_initializable_iterator() | |
# iterator = input_dataset.make_one_shot_iterator() | |
a = iterator.get_next() | |
features = { | |
'id': tf.FixedLenFeature(shape=(), dtype=tf.string), | |
'sequence': tf.FixedLenFeature(shape=(), dtype=tf.string), | |
'audio': tf.FixedLenFeature(shape=(), dtype=tf.string), | |
'velocity_range': tf.FixedLenFeature(shape=(), dtype=tf.string), | |
} | |
record = tf.parse_single_example(a, features) | |
sequence_id, sequence, audio, velocity_range = \ | |
record['id'], record['sequence'], record['audio'], record['velocity_range'] | |
sess = tf.Session() | |
sess.run(iterator.initializer) | |
# Dump wav | |
audio_content = audio.eval(session=sess) | |
with open('test.wav', 'wb') as f: | |
f.write(audio_content) | |
## Dump midi | |
sequence_content = sequence.eval(session=sess) | |
note_sequence = music_pb2.NoteSequence.FromString(sequence_content) | |
midi_filename = ('test_direct.midi') | |
midi_io.sequence_proto_to_midi_file(note_sequence, midi_filename) | |
if __name__ == '__main__': | |
main() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment