def generate_double(fn, gn):
test_piano_roll = midiToPianoroll(fn)
test_data = [test_piano_roll]
test_input = createSeqTestNetInputs(test_data,
cfg.MODEL_PARAMS.X_SEQ_LENGTH)
test_data = test_input[0]
net_output = model.predict(test_data)
net_roll = seqNetOutToPianoroll(net_output)
total_roll = np.concatenate((test_piano_roll, net_roll))
pianorollToMidi(net_roll, gn)
def test():
midi_files = glob.glob(DATA_DIR + '/*.mid')
file_idx = random.randint(0, len(midi_files) - 1)
# midi_file = midi_files[file_idx]
midi_file = 'uploads/mozk175b_trimmed.mid'
print('Generating from %s' % midi_file)
test_piano_roll = midiToPianoroll(midi_file)
test_data = [test_piano_roll]
test_input = createSeqTestNetInputs(test_data,
cfg.MODEL_PARAMS.X_SEQ_LENGTH)
test_data = test_input[0]
generated_file = 'AI_generated_%s.mid' % (
time.strftime("%Y_%m_%d_%H_%M_%s"))
generated_path = '%s/%s' % (cfg.DATA.GENERATED_DIR, generated_file)
net_output = model.predict(test_data)
print("net_output:", np.array(net_output.shape))
net_roll = seqNetOutToPianoroll(net_output)
print("net_roll:", net_roll.shape)
pianorollToMidi(net_roll, generated_path)
def generate():
file = flask.request.files['midifile']
if file.filename.split('.')[-1] in ALLOWED_EXTENSIONS:
print(model.summary())
fn = 'uploads/input_file_%s.mid' % (time.strftime("%Y_%m_%d_%H_%M_%s"))
file.save(fn)
gn = 'static/AI_generated_%s.mid' % (
time.strftime("%Y_%m_%d_%H_%M_%s"))
# zen.generate_double(fn, gn)
test_piano_roll = midiToPianoroll(fn)
test_data = [test_piano_roll]
test_input = createSeqTestNetInputs(test_data,
cfg.MODEL_PARAMS.X_SEQ_LENGTH)
test_data = test_input[0]
net_output = model.predict(test_data)
net_roll = seqNetOutToPianoroll(net_output)
total_roll = np.concatenate((test_piano_roll, net_roll))
pianorollToMidi(net_roll, gn)
return flask.jsonify({'generated': gn})
else:
return "You have uploaded invalid file"
block_blob_service.get_blob_to_path('musicmodels', cfg.DATA.MODEL_FILE,
cfg.DATA.MODEL_PATH)
block_blob_service.get_blob_to_path('musicmodels', cfg.DATA.WEIGHTS_FILE,
cfg.DATA.WEIGHTS_PATH)
model = model_from_json(open(cfg.DATA.MODEL_PATH).read())
model.load_weights(cfg.DATA.WEIGHTS_PATH)
model.compile(loss=cfg.MODEL_PARAMS.LOSS_FUNCTION,
optimizer=cfg.MODEL_PARAMS.OPTIMIZER)
if __name__ == '__main__':
# Import the logger only for Workbench runs
logger = get_azureml_logger()
init()
#PRIMER
dataset_folder = download_grocery_data()
midi_files = glob.glob(dataset_folder + '/*.mid')
#choose a random file as a primer
file_idx = random.randint(0, len(midi_files) - 1)
primer = midi_files[file_idx]
test_piano_roll = midiToPianoroll(primer)
test_data = [test_piano_roll]
test_input = createSeqTestNetInputs(test_data,
cfg.MODEL_PARAMS.X_SEQ_LENGTH)
run(test_input)
def test_identity_fn():
# verify the identity function
midi_file = 'uploads/mozk175b_trimmed.mid'
test_piano_roll = midiToPianoroll(midi_file)
generated_path = '%s/%s' % (cfg.DATA.GENERATED_DIR, 'same.mid')
pianorollToMidi(test_piano_roll, generated_path)
default=0)
args = parser.parse_args()
if args.origin_length <= 0:
print("invalid origin length")
exit()
if args.target_length <= 0:
print("invalid target length")
exit()
if args.load_epoch <= 0:
print("invalid load epoch")
exit()
piano_data = midiToPianoroll(path, debug=True)
print("shape of data ", piano_data.shape)
input_datax = torch.from_numpy(
piano_data[0:args.origin_length, :]).unsqueeze(1).float()
encoder1 = EncoderRNN(input_dim, hidden_dim).to(device)
decoder1 = DecoderRNN(input_dim, hidden_dim).to(device)
encoder1.load_state_dict(
torch.load('../models/encoder_baseline_' + str(args.load_epoch) +
'_Adam1e-3'))
decoder1.load_state_dict(
torch.load('../models/decoder_baseline_' + str(args.load_epoch) +
'_Adam1e-3'))