chore: import upstream snapshot with attribution
This commit is contained in:
wehub-resource-sync
@@ -0,0 +1,18 @@
|
||||
# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License"
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
|
||||
from .esc50 import ESC50
|
||||
from .tess import TESS
|
||||
|
||||
__all__ = ["ESC50", "TESS"]
|
||||
@@ -0,0 +1,98 @@
|
||||
# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
from __future__ import annotations
|
||||
|
||||
import paddle
|
||||
|
||||
from ..features import MFCC, LogMelSpectrogram, MelSpectrogram, Spectrogram
|
||||
|
||||
feat_funcs = {
|
||||
'raw': None,
|
||||
'melspectrogram': MelSpectrogram,
|
||||
'mfcc': MFCC,
|
||||
'logmelspectrogram': LogMelSpectrogram,
|
||||
'spectrogram': Spectrogram,
|
||||
}
|
||||
|
||||
|
||||
class AudioClassificationDataset(paddle.io.Dataset):
|
||||
"""
|
||||
Base class of audio classification dataset.
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
files: list[str],
|
||||
labels: list[int],
|
||||
feat_type: str = 'raw',
|
||||
sample_rate: int | None = None,
|
||||
**kwargs,
|
||||
):
|
||||
"""
|
||||
Args:
|
||||
files (:obj:`List[str]`): A list of absolute path of audio files.
|
||||
labels (:obj:`List[int]`): Labels of audio files.
|
||||
feat_type (:obj:`str`, `optional`, defaults to `raw`):
|
||||
It identifies the feature type that user wants to extract an audio file.
|
||||
"""
|
||||
super().__init__()
|
||||
|
||||
if feat_type not in feat_funcs.keys():
|
||||
raise RuntimeError(
|
||||
f"Unknown feat_type: {feat_type}, it must be one in {list(feat_funcs.keys())}"
|
||||
)
|
||||
|
||||
self.files = files
|
||||
self.labels = labels
|
||||
|
||||
self.feat_type = feat_type
|
||||
self.sample_rate = sample_rate
|
||||
self.feat_config = (
|
||||
kwargs # Pass keyword arguments to customize feature config
|
||||
)
|
||||
|
||||
def _get_data(self, input_file: str):
|
||||
raise NotImplementedError
|
||||
|
||||
def _convert_to_record(self, idx):
|
||||
file, label = self.files[idx], self.labels[idx]
|
||||
waveform, sample_rate = paddle.audio.load(file)
|
||||
self.sample_rate = sample_rate
|
||||
|
||||
feat_func = feat_funcs[self.feat_type]
|
||||
|
||||
record = {}
|
||||
if len(waveform.shape) == 2:
|
||||
waveform = waveform.squeeze(0) # 1D input
|
||||
waveform = paddle.to_tensor(waveform, dtype=paddle.float32)
|
||||
if feat_func is not None:
|
||||
waveform = waveform.unsqueeze(0) # (batch_size, T)
|
||||
if self.feat_type != 'spectrogram':
|
||||
feature_extractor = feat_func(
|
||||
sr=self.sample_rate, **self.feat_config
|
||||
)
|
||||
else:
|
||||
feature_extractor = feat_func(**self.feat_config)
|
||||
record['feat'] = feature_extractor(waveform).squeeze(0)
|
||||
else:
|
||||
record['feat'] = waveform
|
||||
record['label'] = label
|
||||
return record
|
||||
|
||||
def __getitem__(self, idx):
|
||||
record = self._convert_to_record(idx)
|
||||
return record['feat'], record['label']
|
||||
|
||||
def __len__(self):
|
||||
return len(self.files)
|
||||
@@ -0,0 +1,227 @@
|
||||
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
from __future__ import annotations
|
||||
|
||||
import os
|
||||
from typing import TYPE_CHECKING, Any, Literal, NamedTuple, TypeAlias
|
||||
|
||||
from paddle.dataset.common import DATA_HOME
|
||||
from paddle.utils import download
|
||||
|
||||
from .dataset import AudioClassificationDataset
|
||||
|
||||
if TYPE_CHECKING:
|
||||
_ModeLiteral: TypeAlias = Literal[
|
||||
'train',
|
||||
'dev',
|
||||
]
|
||||
_FeatTypeLiteral: TypeAlias = Literal[
|
||||
'raw',
|
||||
'melspectrogram',
|
||||
'mfcc',
|
||||
'logmelspectrogram',
|
||||
'spectrogram',
|
||||
]
|
||||
|
||||
|
||||
__all__ = []
|
||||
|
||||
|
||||
class ESC50(AudioClassificationDataset):
|
||||
"""
|
||||
The ESC-50 dataset is a labeled collection of 2000 environmental audio recordings
|
||||
suitable for benchmarking methods of environmental sound classification. The dataset
|
||||
consists of 5-second-long recordings organized into 50 semantical classes (with
|
||||
40 examples per class)
|
||||
|
||||
Reference:
|
||||
ESC: Dataset for Environmental Sound Classification
|
||||
http://dx.doi.org/10.1145/2733373.2806390
|
||||
|
||||
Args:
|
||||
mode (str, optional): It identifies the dataset mode (train or dev). Default:train.
|
||||
split (int, optional): It specify the fold of dev dataset. Default:1.
|
||||
feat_type (str, optional): It identifies the feature type that user wants to extract of an audio file. Default:raw.
|
||||
archive(dict, optional): it tells where to download the audio archive. Default:None.
|
||||
|
||||
Returns:
|
||||
:ref:`api_paddle_io_Dataset`. An instance of ESC50 dataset.
|
||||
|
||||
Examples:
|
||||
|
||||
.. code-block:: pycon
|
||||
|
||||
>>> # doctest: +TIMEOUT(60)
|
||||
>>> import paddle
|
||||
|
||||
>>> esc50_dataset = paddle.audio.datasets.ESC50(
|
||||
... mode='dev',
|
||||
... feat_type='raw',
|
||||
... )
|
||||
>>> for idx in range(5):
|
||||
... audio, label = esc50_dataset[idx]
|
||||
... # do something with audio, label
|
||||
... print(audio.shape, label)
|
||||
... # [audio_data_length] , label_id
|
||||
paddle.Size([220500]) 0
|
||||
paddle.Size([220500]) 14
|
||||
paddle.Size([220500]) 36
|
||||
paddle.Size([220500]) 36
|
||||
paddle.Size([220500]) 19
|
||||
|
||||
>>> esc50_dataset = paddle.audio.datasets.ESC50(
|
||||
... mode='dev',
|
||||
... feat_type='mfcc',
|
||||
... n_mfcc=40,
|
||||
... )
|
||||
>>> for idx in range(5):
|
||||
... audio, label = esc50_dataset[idx]
|
||||
... # do something with mfcc feature, label
|
||||
... print(audio.shape, label)
|
||||
... # [feature_dim, length] , label_id
|
||||
paddle.Size([40, 1723]) 0
|
||||
paddle.Size([40, 1723]) 14
|
||||
paddle.Size([40, 1723]) 36
|
||||
paddle.Size([40, 1723]) 36
|
||||
paddle.Size([40, 1723]) 19
|
||||
|
||||
"""
|
||||
|
||||
archive: dict[str, str] = {
|
||||
'url': 'https://paddleaudio.bj.bcebos.com/datasets/ESC-50-master.zip',
|
||||
'md5': '7771e4b9d86d0945acce719c7a59305a',
|
||||
}
|
||||
|
||||
label_list: list[str] = [
|
||||
# Animals
|
||||
'Dog',
|
||||
'Rooster',
|
||||
'Pig',
|
||||
'Cow',
|
||||
'Frog',
|
||||
'Cat',
|
||||
'Hen',
|
||||
'Insects (flying)',
|
||||
'Sheep',
|
||||
'Crow',
|
||||
# Natural soundscapes & water sounds
|
||||
'Rain',
|
||||
'Sea waves',
|
||||
'Crackling fire',
|
||||
'Crickets',
|
||||
'Chirping birds',
|
||||
'Water drops',
|
||||
'Wind',
|
||||
'Pouring water',
|
||||
'Toilet flush',
|
||||
'Thunderstorm',
|
||||
# Human, non-speech sounds
|
||||
'Crying baby',
|
||||
'Sneezing',
|
||||
'Clapping',
|
||||
'Breathing',
|
||||
'Coughing',
|
||||
'Footsteps',
|
||||
'Laughing',
|
||||
'Brushing teeth',
|
||||
'Snoring',
|
||||
'Drinking, sipping',
|
||||
# Interior/domestic sounds
|
||||
'Door knock',
|
||||
'Mouse click',
|
||||
'Keyboard typing',
|
||||
'Door, wood creaks',
|
||||
'Can opening',
|
||||
'Washing machine',
|
||||
'Vacuum cleaner',
|
||||
'Clock alarm',
|
||||
'Clock tick',
|
||||
'Glass breaking',
|
||||
# Exterior/urban noises
|
||||
'Helicopter',
|
||||
'Chainsaw',
|
||||
'Siren',
|
||||
'Car horn',
|
||||
'Engine',
|
||||
'Train',
|
||||
'Church bells',
|
||||
'Airplane',
|
||||
'Fireworks',
|
||||
'Hand saw',
|
||||
]
|
||||
meta: str = os.path.join('ESC-50-master', 'meta', 'esc50.csv')
|
||||
audio_path: str = os.path.join('ESC-50-master', 'audio')
|
||||
|
||||
class meta_info(NamedTuple):
|
||||
filename: str
|
||||
fold: str
|
||||
target: str
|
||||
category: str
|
||||
esc10: str
|
||||
src_file: str
|
||||
take: str
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
mode: _ModeLiteral = 'train',
|
||||
split: int = 1,
|
||||
feat_type: _FeatTypeLiteral = 'raw',
|
||||
archive: dict[str, str] | None = None,
|
||||
**kwargs: Any,
|
||||
) -> None:
|
||||
assert split in range(1, 6), (
|
||||
f'The selected split should be integer, and 1 <= split <= 5, but got {split}'
|
||||
)
|
||||
if archive is not None:
|
||||
self.archive = archive
|
||||
files, labels = self._get_data(mode, split)
|
||||
super().__init__(
|
||||
files=files, labels=labels, feat_type=feat_type, **kwargs
|
||||
)
|
||||
|
||||
def _get_meta_info(self) -> list[meta_info]:
|
||||
ret = []
|
||||
with open(os.path.join(DATA_HOME, self.meta), 'r') as rf:
|
||||
for line in rf.readlines()[1:]:
|
||||
ret.append(self.meta_info(*line.strip().split(',')))
|
||||
return ret
|
||||
|
||||
def _get_data(
|
||||
self, mode: _ModeLiteral, split: int
|
||||
) -> tuple[list[str], list[int]]:
|
||||
if not os.path.isdir(
|
||||
os.path.join(DATA_HOME, self.audio_path)
|
||||
) or not os.path.isfile(os.path.join(DATA_HOME, self.meta)):
|
||||
download.get_path_from_url(
|
||||
self.archive['url'],
|
||||
DATA_HOME,
|
||||
self.archive['md5'],
|
||||
decompress=True,
|
||||
)
|
||||
|
||||
meta_info = self._get_meta_info()
|
||||
|
||||
files = []
|
||||
labels = []
|
||||
for sample in meta_info:
|
||||
filename, fold, target, _, _, _, _ = sample
|
||||
if mode == 'train' and int(fold) != split:
|
||||
files.append(os.path.join(DATA_HOME, self.audio_path, filename))
|
||||
labels.append(int(target))
|
||||
|
||||
if mode != 'train' and int(fold) == split:
|
||||
files.append(os.path.join(DATA_HOME, self.audio_path, filename))
|
||||
labels.append(int(target))
|
||||
|
||||
return files, labels
|
||||
@@ -0,0 +1,166 @@
|
||||
# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
from __future__ import annotations
|
||||
|
||||
import os
|
||||
from typing import TYPE_CHECKING, Any, NamedTuple
|
||||
|
||||
from paddle.dataset.common import DATA_HOME
|
||||
from paddle.utils import download
|
||||
|
||||
from .dataset import AudioClassificationDataset
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from .esc50 import _FeatTypeLiteral, _ModeLiteral
|
||||
|
||||
__all__ = []
|
||||
|
||||
|
||||
class TESS(AudioClassificationDataset):
|
||||
"""
|
||||
TESS is a set of 200 target words were spoken in the carrier phrase
|
||||
"Say the word _____' by two actresses (aged 26 and 64 years) and
|
||||
recordings were made of the set portraying each of seven emotions(anger,
|
||||
disgust, fear, happiness, pleasant surprise, sadness, and neutral).
|
||||
There are 2800 stimuli in total.
|
||||
|
||||
Reference:
|
||||
Toronto emotional speech set (TESS) https://tspace.library.utoronto.ca/handle/1807/24487
|
||||
https://doi.org/10.5683/SP2/E8H2MF
|
||||
|
||||
Args:
|
||||
mode (str, optional): It identifies the dataset mode (train or dev). Defaults to train.
|
||||
n_folds (int, optional): Split the dataset into n folds. 1 fold for dev dataset and n-1 for train dataset. Defaults to 5.
|
||||
split (int, optional): It specify the fold of dev dataset. Defaults to 1.
|
||||
feat_type (str, optional): It identifies the feature type that user wants to extract of an audio file. Defaults to raw.
|
||||
archive(dict): it tells where to download the audio archive. Defaults to None.
|
||||
|
||||
Returns:
|
||||
:ref:`api_paddle_io_Dataset`. An instance of TESS dataset.
|
||||
|
||||
Examples:
|
||||
|
||||
.. code-block:: pycon
|
||||
|
||||
>>> # doctest: +TIMEOUT(60)
|
||||
>>> import paddle
|
||||
|
||||
>>> tess_dataset = paddle.audio.datasets.TESS(
|
||||
... mode='dev',
|
||||
... feat_type='raw',
|
||||
... )
|
||||
>>> for idx in range(5):
|
||||
... audio, label = tess_dataset[idx]
|
||||
... # do something with audio, label
|
||||
... print(audio.shape, label)
|
||||
... # [audio_data_length] , label_id
|
||||
|
||||
>>> tess_dataset = paddle.audio.datasets.TESS(
|
||||
... mode='dev',
|
||||
... feat_type='mfcc',
|
||||
... n_mfcc=40,
|
||||
... )
|
||||
>>> for idx in range(5):
|
||||
... audio, label = tess_dataset[idx]
|
||||
... # do something with mfcc feature, label
|
||||
... print(audio.shape, label)
|
||||
... # [feature_dim, num_frames] , label_id
|
||||
"""
|
||||
|
||||
archive: dict[str, str] = {
|
||||
'url': 'https://bj.bcebos.com/paddleaudio/datasets/TESS_Toronto_emotional_speech_set.zip',
|
||||
'md5': '1465311b24d1de704c4c63e4ccc470c7',
|
||||
}
|
||||
|
||||
label_list: list[str] = [
|
||||
'angry',
|
||||
'disgust',
|
||||
'fear',
|
||||
'happy',
|
||||
'neutral',
|
||||
'ps', # pleasant surprise
|
||||
'sad',
|
||||
]
|
||||
|
||||
audio_path: str = 'TESS_Toronto_emotional_speech_set'
|
||||
|
||||
class meta_info(NamedTuple):
|
||||
speaker: str
|
||||
word: str
|
||||
emotion: str
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
mode: _ModeLiteral = 'train',
|
||||
n_folds: int = 5,
|
||||
split: int = 1,
|
||||
feat_type: _FeatTypeLiteral = 'raw',
|
||||
archive: dict[str, str] | None = None,
|
||||
**kwargs: Any,
|
||||
) -> None:
|
||||
assert isinstance(n_folds, int) and (n_folds >= 1), (
|
||||
f'the n_folds should be integer and n_folds >= 1, but got {n_folds}'
|
||||
)
|
||||
assert split in range(1, n_folds + 1), (
|
||||
f'The selected split should be integer and should be 1 <= split <= {n_folds}, but got {split}'
|
||||
)
|
||||
if archive is not None:
|
||||
self.archive = archive
|
||||
files, labels = self._get_data(mode, n_folds, split)
|
||||
super().__init__(
|
||||
files=files, labels=labels, feat_type=feat_type, **kwargs
|
||||
)
|
||||
|
||||
def _get_meta_info(self, files) -> list[meta_info]:
|
||||
ret = []
|
||||
for file in files:
|
||||
basename_without_extend = os.path.basename(file)[:-4]
|
||||
ret.append(self.meta_info(*basename_without_extend.split('_')))
|
||||
return ret
|
||||
|
||||
def _get_data(
|
||||
self, mode: str, n_folds: int, split: int
|
||||
) -> tuple[list[str], list[int]]:
|
||||
if not os.path.isdir(os.path.join(DATA_HOME, self.audio_path)):
|
||||
download.get_path_from_url(
|
||||
self.archive['url'],
|
||||
DATA_HOME,
|
||||
self.archive['md5'],
|
||||
decompress=True,
|
||||
)
|
||||
|
||||
wav_files = []
|
||||
for root, _, files in os.walk(os.path.join(DATA_HOME, self.audio_path)):
|
||||
for file in files:
|
||||
if file.endswith('.wav'):
|
||||
wav_files.append(os.path.join(root, file))
|
||||
|
||||
meta_info = self._get_meta_info(wav_files)
|
||||
|
||||
files = []
|
||||
labels = []
|
||||
for idx, sample in enumerate(meta_info):
|
||||
_, _, emotion = sample
|
||||
target = self.label_list.index(emotion)
|
||||
fold = idx % n_folds + 1
|
||||
|
||||
if mode == 'train' and int(fold) != split:
|
||||
files.append(wav_files[idx])
|
||||
labels.append(target)
|
||||
|
||||
if mode != 'train' and int(fold) == split:
|
||||
files.append(wav_files[idx])
|
||||
labels.append(target)
|
||||
|
||||
return files, labels
|
||||
Reference in New Issue
Block a user