Spaces:
Running
on
A10G
Running
on
A10G
File size: 4,459 Bytes
0883aa1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 |
# Copyright (c) 2023 Amphion.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
import os
import json
import librosa
from tqdm import tqdm
from collections import defaultdict
from utils.util import has_existed
from preprocessors import GOLDEN_TEST_SAMPLES
def get_test_songs():
golden_samples = GOLDEN_TEST_SAMPLES["m4singer"]
# every item is a tuple (singer, song)
golden_songs = [s.split("_")[:2] for s in golden_samples]
# singer_song, eg: Alto-1_美错
golden_songs = ["_".join(t) for t in golden_songs]
return golden_songs
def m4singer_statistics(meta):
singers = []
songs = []
singer2songs = defaultdict(lambda: defaultdict(list))
for utt in meta:
p, s, uid = utt["item_name"].split("#")
singers.append(p)
songs.append(s)
singer2songs[p][s].append(uid)
unique_singers = list(set(singers))
unique_songs = list(set(songs))
unique_singers.sort()
unique_songs.sort()
print(
"M4Singer: {} singers, {} utterances ({} unique songs)".format(
len(unique_singers), len(songs), len(unique_songs)
)
)
print("Singers: \n{}".format("\t".join(unique_singers)))
return singer2songs, unique_singers
def main(output_path, dataset_path):
print("-" * 10)
print("Preparing test samples for m4singer...\n")
save_dir = os.path.join(output_path, "m4singer")
os.makedirs(save_dir, exist_ok=True)
train_output_file = os.path.join(save_dir, "train.json")
test_output_file = os.path.join(save_dir, "test.json")
singer_dict_file = os.path.join(save_dir, "singers.json")
utt2singer_file = os.path.join(save_dir, "utt2singer")
if (
has_existed(train_output_file)
and has_existed(test_output_file)
and has_existed(singer_dict_file)
and has_existed(utt2singer_file)
):
return
utt2singer = open(utt2singer_file, "w")
# Load
m4singer_dir = dataset_path
meta_file = os.path.join(m4singer_dir, "meta.json")
with open(meta_file, "r", encoding="utf-8") as f:
meta = json.load(f)
singer2songs, unique_singers = m4singer_statistics(meta)
test_songs = get_test_songs()
# We select songs of standard samples as test songs
train = []
test = []
train_index_count = 0
test_index_count = 0
train_total_duration = 0
test_total_duration = 0
for singer, songs in tqdm(singer2songs.items()):
song_names = list(songs.keys())
for chosen_song in song_names:
chosen_song = chosen_song.replace(" ", "-")
for chosen_uid in songs[chosen_song]:
res = {
"Dataset": "m4singer",
"Singer": singer,
"Song": chosen_song,
"Uid": "{}_{}_{}".format(singer, chosen_song, chosen_uid),
}
res["Path"] = os.path.join(
m4singer_dir, "{}#{}/{}.wav".format(singer, chosen_song, chosen_uid)
)
assert os.path.exists(res["Path"])
duration = librosa.get_duration(filename=res["Path"])
res["Duration"] = duration
if "_".join([singer, chosen_song]) in test_songs:
res["index"] = test_index_count
test_total_duration += duration
test.append(res)
test_index_count += 1
else:
res["index"] = train_index_count
train_total_duration += duration
train.append(res)
train_index_count += 1
utt2singer.write("{}\t{}\n".format(res["Uid"], res["Singer"]))
print("#Train = {}, #Test = {}".format(len(train), len(test)))
print(
"#Train hours= {}, #Test hours= {}".format(
train_total_duration / 3600, test_total_duration / 3600
)
)
# Save train.json and test.json
with open(train_output_file, "w") as f:
json.dump(train, f, indent=4, ensure_ascii=False)
with open(test_output_file, "w") as f:
json.dump(test, f, indent=4, ensure_ascii=False)
# Save singers.json
singer_lut = {name: i for i, name in enumerate(unique_singers)}
with open(singer_dict_file, "w") as f:
json.dump(singer_lut, f, indent=4, ensure_ascii=False)
|