def main(mode, src, dst, word_vectors_path, embedding_dict_path, transpose):
to_numpy = mode == AVAILABLE_DIRECTIONS[0]
input_paths, output_paths = parse_file_paths(src, dst, to_numpy, '.npy')
print(input_paths, output_paths)
wv = KeyedVectors.load(word_vectors_path, mmap='r')
with open(embedding_dict_path) as fp:
embedding_dict = json.load(fp)
for input_p, output_p in zip(input_paths, output_paths):
if to_numpy:
mid = MidiFile(input_p)
if transpose:
for step in range(-12, 13):
transposed_mid = transpose_midi(step, mid)
numpy_track = mid2np(transposed_mid, wv, embedding_dict)
save_numpy_midi(output_p.replace(
'.npy', f'_{step}.npy'), numpy_track)
else:
numpy_track = mid2np(mid, wv, embedding_dict)
save_numpy_midi(output_p, numpy_track)
else:
numpy_track = read_numpy_midi(input_p)
mid = np2mid(numpy_track, wv)
mid.save(output_p)
def main(src, dst, low_step, high_step):
to_numpy = True
input_paths, output_paths = parse_file_paths(src, dst, to_numpy)
for input_p, output_p in zip(input_paths, output_paths):
track = read_numpy_midi(input_p)
for step in range(low_step, high_step):
output_path = output_p.replace('.', f'_tr_{str(step)}.')
transposed = transpose(track, step)
save_numpy_midi(output_path, transposed)
def main(src, dst, transposition, **kwargs):
input_paths, output_paths = parse_file_paths(src, dst, True)
for input_p, output_p in zip(input_paths, output_paths):
mid = MidiFile(input_p)
numpy_track = just_sparsify(mid)
if not transposition:
save_numpy_midi(output_p, numpy_track)
else:
for step in range(-12, 13):
output_path = output_p.replace('.', f'_tr_{str(step)}.')
transposed = transpose(numpy_track, step)
save_numpy_midi(output_path, transposed)
def main(mode, src, dst, **kwargs):
clean_kwargs = {k: v for k, v in kwargs.items() if v is not None}
to_numpy = mode == AVAILABLE_DIRECTIONS[0]
input_paths, output_paths = parse_file_paths(src, dst, to_numpy)
for input_p, output_p in zip(input_paths, output_paths):
if to_numpy:
mid = MidiFile(input_p)
messages = [m for m in mid]
numpy_track = mid2np(messages, **clean_kwargs)
save_numpy_midi(output_p, numpy_track)
else:
numpy_track = read_numpy_midi(input_p)
mid = np2mid(numpy_track, **clean_kwargs)
mid.save(output_p)
def main(src, limit_vector, **kwargs):
clean_kwargs = {k: v for k, v in kwargs.items() if v is not None}
files_in_folder = get_valid_files_in_dir(src)
file_paths = [os.path.join(src, f) for f in files_in_folder]
tracks = [sparse.load_npz(fp).tocsr() for fp in file_paths]
if limit_vector:
print(f'limiting to {limit_vector}')
tracks = [t[:, :limit_vector] for t in tracks]
counter, counter_notes = create_counters(tracks)
embedding_dict = create_dict(counter_notes, **clean_kwargs)
dicted_dataset, dicted_dataset_subsequents_removed = dictify_dataset(
tracks, embedding_dict)
output_dir = os.path.join(src, 'meta')
os.makedirs(output_dir, exist_ok=True)
options_str = option_kwargs_to_string(clean_kwargs)
with open(os.path.join(output_dir, f'_counter_nums_{options_str}.json'),
'w') as f:
json.dump(counter, f, indent=4)
with open(os.path.join(output_dir, f'_counter_notes_{options_str}.json'),
'w') as f:
json.dump(counter_notes, f, indent=4)
with open(os.path.join(output_dir, f'_embedding_dict_{options_str}.json'),
'w') as f:
json.dump(embedding_dict, f, indent=4)
save_numpy_midi(
os.path.join(output_dir, f'_dicted_dataset_{options_str}.npy'),
dicted_dataset)
save_numpy_midi(
os.path.join(output_dir,
f'_dicted_dataset_subsequents_removed_{options_str}.npy'),
dicted_dataset_subsequents_removed)
def main(mode, src, dst, dur_dict, do_transpose, **kwargs):
clean_kwargs = {k: v for k, v in kwargs.items() if v is not None}
to_numpy = mode == AVAILABLE_DIRECTIONS[0]
input_paths, output_paths = parse_file_paths(src, dst, to_numpy)
if to_numpy:
tracks, durations = zip(
*[mid2np(MidiFile(input_p), **clean_kwargs) for input_p in input_paths])
tracks_blob = np.concatenate(tracks, axis=0)
durations = np.concatenate(durations)
durations_ohe, _, durations_dict = cluster_and_ohe_durations(
durations.reshape(-1, 1))
meta_dir = os.path.join(dst, 'meta')
os.makedirs(meta_dir, exist_ok=True)
with open(os.path.join(meta_dir, 'durations_dict.json'), 'w+') as fp:
json.dump(durations_dict, fp, indent=4, default=lambda x: float(
x) if isinstance(x, np.float32) else x)
track_w_duration_blob = np.concatenate(
(tracks_blob, durations_ohe), axis=1)
save_numpy_midi(os.path.join(
meta_dir, 'tracks_blob_w_duration.npz'), track_w_duration_blob)
track_lens = [t.shape[0] for t in tracks]
durations_ohe_split = np.split(
durations_ohe, np.cumsum(track_lens[:-1]))
for output_p, t, d in zip(output_paths, tracks, durations_ohe_split):
if do_transpose:
for step in range(-12, 13):
transposed_t = transpose(t, step)
track_w_dur = np.concatenate((transposed_t, d), axis=-1)
save_numpy_midi(os.path.join(output_p.replace(
'.npz', f'_{step}.npz')), track_w_dur)
else:
track_w_dur = np.concatenate((t, d), axis=-1)
save_numpy_midi(os.path.join(output_p), track_w_dur)
else:
meta_dir = os.path.join(dst, 'meta')
assert dur_dict != '', 'Please specify duration dict'
with open(dur_dict) as fp:
durations_dict = json.load(fp)
for input_p, output_p in zip(input_paths, output_paths):
numpy_track = read_numpy_midi(input_p)
track, durations = np.split(numpy_track, [128], axis=1)
mid = np2mid(track, durations, durations_dict, **clean_kwargs)
mid.save(output_p)