def test_prepare_tfrecord_no_f0_and_loudness(self):
sample_rate = 16000
prepare_tfrecord_lib.prepare_tfrecord(
[self.wav_path],
os.path.join(self.test_dir, 'output.tfrecord'),
num_shards=2,
sample_rate=sample_rate,
frame_rate=None,
window_secs=None)
self.validate_outputs(1, {'audio': self.wav_secs * sample_rate})
def test_prepare_tfrecord_at_indivisible_sample_rate_throws_error(
self, sample_rate):
frame_rate = 250
with self.assertRaises(ValueError):
prepare_tfrecord_lib.prepare_tfrecord([self.wav_path],
os.path.join(
self.test_dir,
'output.tfrecord'),
num_shards=2,
sample_rate=sample_rate,
frame_rate=frame_rate,
window_secs=None)
def run():
input_audio_paths = []
for filepattern in FLAGS.input_audio_filepatterns:
input_audio_paths.extend(tf.io.gfile.glob(filepattern))
prepare_tfrecord(input_audio_paths,
FLAGS.output_tfrecord_path,
num_shards=FLAGS.num_shards,
sample_rate=FLAGS.sample_rate,
frame_rate=FLAGS.frame_rate,
window_secs=FLAGS.example_secs,
hop_secs=FLAGS.sliding_window_hop_secs,
pipeline_options=FLAGS.pipeline_options)
def test_prepare_tfrecord_no_f0_and_loudness(self, sample_rate):
prepare_tfrecord_lib.prepare_tfrecord(
[self.wav_path],
os.path.join(self.test_dir, 'output.tfrecord'),
num_shards=2,
sample_rate=sample_rate,
frame_rate=None,
window_secs=None)
self.validate_outputs(
1, {
'audio': int(self.wav_secs * sample_rate),
'audio_crepe': int(self.wav_secs * _CREPE_SAMPLE_RATE)
})
def buildTFRecords(audio_input, output_tfrecord_path):
logging.info("Building TFRecords")
if not glob.glob(audio_input):
raise ValueError('No audio files found. Please use the previous cell to '
'upload.')
else:
logging.info("found " + audio_input)
input_audio_paths = []
input_audio_paths.extend(tf.io.gfile.glob(audio_input))
prepare_tfrecord(
input_audio_paths,
output_tfrecord_path,
num_shards=10,
pipeline_options='--runner=DirectRunner')
def test_prepare_tfrecord_no_split(self, sample_rate):
frame_rate = 250
prepare_tfrecord_lib.prepare_tfrecord(
[self.wav_path],
os.path.join(self.test_dir, 'output.tfrecord'),
num_shards=2,
sample_rate=sample_rate,
frame_rate=frame_rate,
window_secs=None)
expected_f0_and_loudness_length = int(self.wav_secs * frame_rate)
self.validate_outputs(
1, {
'audio': int(self.wav_secs * sample_rate),
'f0_hz': expected_f0_and_loudness_length,
'f0_confidence': expected_f0_and_loudness_length,
'loudness_db': expected_f0_and_loudness_length,
})
def test_prepare_tfrecord(self, sample_rate):
frame_rate = 250
window_secs = 1
hop_secs = 0.5
prepare_tfrecord_lib.prepare_tfrecord(
[self.wav_path],
os.path.join(self.test_dir, 'output.tfrecord'),
num_shards=2,
sample_rate=sample_rate,
frame_rate=frame_rate,
window_secs=window_secs,
hop_secs=hop_secs)
expected_f0_and_loudness_length = int(window_secs * frame_rate)
self.validate_outputs(
4, {
'audio': window_secs * sample_rate,
'f0_hz': expected_f0_and_loudness_length,
'f0_confidence': expected_f0_and_loudness_length,
'loudness_db': expected_f0_and_loudness_length,
})
def run():
input_audio_paths = []
for filepattern in FLAGS.input_audio_filepatterns:
input_audio_paths.extend(tf.io.gfile.glob(filepattern))
midi_notes = None
if FLAGS.midi_notes:
notes = FLAGS.midi_notes.split(',')
note_from = int(notes[0])
note_to = int(notes[-1])
midi_notes = list(range(note_from, note_to))
# window_secs is equal to hop_secs, as we do not want overlapping windows
prepare_tfrecord(
input_audio_paths,
FLAGS.output_tfrecord_path,
num_shards=FLAGS.num_shards,
sample_rate=FLAGS.sample_rate,
frame_rate=FLAGS.frame_rate,
window_secs=FLAGS.example_secs,
hop_secs=FLAGS.example_secs,
midi_notes=midi_notes,
pipeline_options=FLAGS.pipeline_options)
