Spaces:
Runtime error
Runtime error
File size: 11,453 Bytes
2b7bf83 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 |
# -*- coding: utf-8 -*-
# Copyright 2019 Tomoki Hayashi
# MIT License (https://opensource.org/licenses/MIT)
"""Dataset modules based on kaldi-style scp files."""
import logging
from multiprocessing import Manager
import kaldiio
import numpy as np
from torch.utils.data import Dataset
from parallel_wavegan.utils import HDF5ScpLoader
from parallel_wavegan.utils import NpyScpLoader
def _get_feats_scp_loader(feats_scp):
# read the first line of feats.scp file
with open(feats_scp) as f:
key, value = f.readlines()[0].replace("\n", "").split()
# check scp type
if ":" in value:
value_1, value_2 = value.split(":")
if value_1.endswith(".ark"):
# kaldi-ark case: utt_id_1 /path/to/utt_id_1.ark:index
return kaldiio.load_scp(feats_scp)
elif value_1.endswith(".h5"):
# hdf5 case with path in hdf5: utt_id_1 /path/to/utt_id_1.h5:feats
return HDF5ScpLoader(feats_scp)
else:
raise ValueError("Not supported feats.scp type.")
else:
if value.endswith(".h5"):
# hdf5 case without path in hdf5: utt_id_1 /path/to/utt_id_1.h5
return HDF5ScpLoader(feats_scp)
elif value.endswith(".npy"):
# npy case: utt_id_1 /path/to/utt_id_1.npy
return NpyScpLoader(feats_scp)
else:
raise ValueError("Not supported feats.scp type.")
class AudioMelSCPDataset(Dataset):
"""PyTorch compatible audio and mel dataset based on kaldi-stype scp files."""
def __init__(
self,
wav_scp,
feats_scp,
segments=None,
audio_length_threshold=None,
mel_length_threshold=None,
return_utt_id=False,
return_sampling_rate=False,
allow_cache=False,
):
"""Initialize dataset.
Args:
wav_scp (str): Kaldi-style wav.scp file.
feats_scp (str): Kaldi-style fests.scp file.
segments (str): Kaldi-style segments file.
audio_length_threshold (int): Threshold to remove short audio files.
mel_length_threshold (int): Threshold to remove short feature files.
return_utt_id (bool): Whether to return utterance id.
return_sampling_rate (bool): Wheter to return sampling rate.
allow_cache (bool): Whether to allow cache of the loaded files.
"""
# load scp as lazy dict
audio_loader = kaldiio.load_scp(wav_scp, segments=segments)
mel_loader = _get_feats_scp_loader(feats_scp)
audio_keys = list(audio_loader.keys())
mel_keys = list(mel_loader.keys())
# filter by threshold
if audio_length_threshold is not None:
audio_lengths = [audio.shape[0] for _, audio in audio_loader.values()]
idxs = [
idx
for idx in range(len(audio_keys))
if audio_lengths[idx] > audio_length_threshold
]
if len(audio_keys) != len(idxs):
logging.warning(
f"Some files are filtered by audio length threshold "
f"({len(audio_keys)} -> {len(idxs)})."
)
audio_keys = [audio_keys[idx] for idx in idxs]
mel_keys = [mel_keys[idx] for idx in idxs]
if mel_length_threshold is not None:
mel_lengths = [mel.shape[0] for mel in mel_loader.values()]
idxs = [
idx
for idx in range(len(mel_keys))
if mel_lengths[idx] > mel_length_threshold
]
if len(mel_keys) != len(idxs):
logging.warning(
f"Some files are filtered by mel length threshold "
f"({len(mel_keys)} -> {len(idxs)})."
)
audio_keys = [audio_keys[idx] for idx in idxs]
mel_keys = [mel_keys[idx] for idx in idxs]
# assert the number of files
assert len(audio_keys) == len(
mel_keys
), f"Number of audio and mel files are different ({len(audio_keys)} vs {len(mel_keys)})."
self.audio_loader = audio_loader
self.mel_loader = mel_loader
self.utt_ids = audio_keys
self.return_utt_id = return_utt_id
self.return_sampling_rate = return_sampling_rate
self.allow_cache = allow_cache
if allow_cache:
# NOTE(kan-bayashi): Manager is need to share memory in dataloader with num_workers > 0
self.manager = Manager()
self.caches = self.manager.list()
self.caches += [() for _ in range(len(self.utt_ids))]
def __getitem__(self, idx):
"""Get specified idx items.
Args:
idx (int): Index of the item.
Returns:
str: Utterance id (only in return_utt_id = True).
ndarray or tuple: Audio signal (T,) or (w/ sampling rate if return_sampling_rate = True).
ndarray: Feature (T', C).
"""
if self.allow_cache and len(self.caches[idx]) != 0:
return self.caches[idx]
utt_id = self.utt_ids[idx]
fs, audio = self.audio_loader[utt_id]
mel = self.mel_loader[utt_id]
# normalize audio signal to be [-1, 1]
audio = audio.astype(np.float32)
audio /= 1 << (16 - 1) # assume that wav is PCM 16 bit
if self.return_sampling_rate:
audio = (audio, fs)
if self.return_utt_id:
items = utt_id, audio, mel
else:
items = audio, mel
if self.allow_cache:
self.caches[idx] = items
return items
def __len__(self):
"""Return dataset length.
Returns:
int: The length of dataset.
"""
return len(self.utt_ids)
class AudioSCPDataset(Dataset):
"""PyTorch compatible audio dataset based on kaldi-stype scp files."""
def __init__(
self,
wav_scp,
segments=None,
audio_length_threshold=None,
return_utt_id=False,
return_sampling_rate=False,
allow_cache=False,
):
"""Initialize dataset.
Args:
wav_scp (str): Kaldi-style wav.scp file.
segments (str): Kaldi-style segments file.
audio_length_threshold (int): Threshold to remove short audio files.
return_utt_id (bool): Whether to return utterance id.
return_sampling_rate (bool): Wheter to return sampling rate.
allow_cache (bool): Whether to allow cache of the loaded files.
"""
# load scp as lazy dict
audio_loader = kaldiio.load_scp(wav_scp, segments=segments)
audio_keys = list(audio_loader.keys())
# filter by threshold
if audio_length_threshold is not None:
audio_lengths = [audio.shape[0] for _, audio in audio_loader.values()]
idxs = [
idx
for idx in range(len(audio_keys))
if audio_lengths[idx] > audio_length_threshold
]
if len(audio_keys) != len(idxs):
logging.warning(
f"Some files are filtered by audio length threshold "
f"({len(audio_keys)} -> {len(idxs)})."
)
audio_keys = [audio_keys[idx] for idx in idxs]
self.audio_loader = audio_loader
self.utt_ids = audio_keys
self.return_utt_id = return_utt_id
self.return_sampling_rate = return_sampling_rate
self.allow_cache = allow_cache
if allow_cache:
# NOTE(kan-bayashi): Manager is need to share memory in dataloader with num_workers > 0
self.manager = Manager()
self.caches = self.manager.list()
self.caches += [() for _ in range(len(self.utt_ids))]
def __getitem__(self, idx):
"""Get specified idx items.
Args:
idx (int): Index of the item.
Returns:
str: Utterance id (only in return_utt_id = True).
ndarray or tuple: Audio signal (T,) or (w/ sampling rate if return_sampling_rate = True).
"""
if self.allow_cache and len(self.caches[idx]) != 0:
return self.caches[idx]
utt_id = self.utt_ids[idx]
fs, audio = self.audio_loader[utt_id]
# normalize audio signal to be [-1, 1]
audio = audio.astype(np.float32)
audio /= 1 << (16 - 1) # assume that wav is PCM 16 bit
if self.return_sampling_rate:
audio = (audio, fs)
if self.return_utt_id:
items = utt_id, audio
else:
items = audio
if self.allow_cache:
self.caches[idx] = items
return items
def __len__(self):
"""Return dataset length.
Returns:
int: The length of dataset.
"""
return len(self.utt_ids)
class MelSCPDataset(Dataset):
"""PyTorch compatible mel dataset based on kaldi-stype scp files."""
def __init__(
self,
feats_scp,
mel_length_threshold=None,
return_utt_id=False,
allow_cache=False,
):
"""Initialize dataset.
Args:
feats_scp (str): Kaldi-style fests.scp file.
mel_length_threshold (int): Threshold to remove short feature files.
return_utt_id (bool): Whether to return utterance id.
allow_cache (bool): Whether to allow cache of the loaded files.
"""
# load scp as lazy dict
mel_loader = _get_feats_scp_loader(feats_scp)
mel_keys = list(mel_loader.keys())
# filter by threshold
if mel_length_threshold is not None:
mel_lengths = [mel.shape[0] for mel in mel_loader.values()]
idxs = [
idx
for idx in range(len(mel_keys))
if mel_lengths[idx] > mel_length_threshold
]
if len(mel_keys) != len(idxs):
logging.warning(
f"Some files are filtered by mel length threshold "
f"({len(mel_keys)} -> {len(idxs)})."
)
mel_keys = [mel_keys[idx] for idx in idxs]
self.mel_loader = mel_loader
self.utt_ids = mel_keys
self.return_utt_id = return_utt_id
self.allow_cache = allow_cache
if allow_cache:
# NOTE(kan-bayashi): Manager is need to share memory in dataloader with num_workers > 0
self.manager = Manager()
self.caches = self.manager.list()
self.caches += [() for _ in range(len(self.utt_ids))]
def __getitem__(self, idx):
"""Get specified idx items.
Args:
idx (int): Index of the item.
Returns:
str: Utterance id (only in return_utt_id = True).
ndarray: Feature (T', C).
"""
if self.allow_cache and len(self.caches[idx]) != 0:
return self.caches[idx]
utt_id = self.utt_ids[idx]
mel = self.mel_loader[utt_id]
if self.return_utt_id:
items = utt_id, mel
else:
items = mel
if self.allow_cache:
self.caches[idx] = items
return items
def __len__(self):
"""Return dataset length.
Returns:
int: The length of dataset.
"""
return len(self.utt_ids)
|