|
|
|
@ -11,18 +11,38 @@
|
|
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
# See the License for the specific language governing permissions and
|
|
|
|
# See the License for the specific language governing permissions and
|
|
|
|
# limitations under the License.
|
|
|
|
# limitations under the License.
|
|
|
|
import collections
|
|
|
|
from dataclasses import dataclass
|
|
|
|
|
|
|
|
from dataclasses import fields
|
|
|
|
from paddle.io import Dataset
|
|
|
|
from paddle.io import Dataset
|
|
|
|
|
|
|
|
|
|
|
|
from paddleaudio import load as load_audio
|
|
|
|
from paddleaudio import load as load_audio
|
|
|
|
|
|
|
|
from paddlespeech.s2t.utils.log import Log
|
|
|
|
|
|
|
|
logger = Log(__name__).getlog()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# the audio meta info in the vector CSVDataset
|
|
|
|
|
|
|
|
# utt_id: the utterance segment name
|
|
|
|
|
|
|
|
# duration: utterance segment time
|
|
|
|
|
|
|
|
# wav: utterance file path
|
|
|
|
|
|
|
|
# start: start point in the original wav file
|
|
|
|
|
|
|
|
# stop: stop point in the original wav file
|
|
|
|
|
|
|
|
# lab_id: the utterance segment's label id
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@dataclass
|
|
|
|
|
|
|
|
class meta_info:
|
|
|
|
|
|
|
|
utt_id: str
|
|
|
|
|
|
|
|
duration: float
|
|
|
|
|
|
|
|
wav: str
|
|
|
|
|
|
|
|
start: int
|
|
|
|
|
|
|
|
stop: int
|
|
|
|
|
|
|
|
lab_id: str
|
|
|
|
|
|
|
|
|
|
|
|
class VoxCelebDataset(Dataset):
|
|
|
|
|
|
|
|
meta_info = collections.namedtuple(
|
|
|
|
|
|
|
|
'META_INFO', ('id', 'duration', 'wav', 'start', 'stop', 'spk_id'))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def __init__(self, csv_path, spk_id2label_path, config):
|
|
|
|
class CSVDataset(Dataset):
|
|
|
|
|
|
|
|
# meta_info = collections.namedtuple(
|
|
|
|
|
|
|
|
# 'META_INFO', ('id', 'duration', 'wav', 'start', 'stop', 'spk_id'))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def __init__(self, csv_path, spk_id2label_path=None, config=None):
|
|
|
|
super().__init__()
|
|
|
|
super().__init__()
|
|
|
|
self.csv_path = csv_path
|
|
|
|
self.csv_path = csv_path
|
|
|
|
self.spk_id2label_path = spk_id2label_path
|
|
|
|
self.spk_id2label_path = spk_id2label_path
|
|
|
|
@ -32,34 +52,41 @@ class VoxCelebDataset(Dataset):
|
|
|
|
|
|
|
|
|
|
|
|
def load_data_csv(self):
|
|
|
|
def load_data_csv(self):
|
|
|
|
data = []
|
|
|
|
data = []
|
|
|
|
|
|
|
|
|
|
|
|
with open(self.csv_path, 'r') as rf:
|
|
|
|
with open(self.csv_path, 'r') as rf:
|
|
|
|
for line in rf.readlines()[1:]:
|
|
|
|
for line in rf.readlines()[1:]:
|
|
|
|
audio_id, duration, wav, start, stop, spk_id = line.strip(
|
|
|
|
audio_id, duration, wav, start, stop, spk_id = line.strip(
|
|
|
|
).split(',')
|
|
|
|
).split(',')
|
|
|
|
data.append(
|
|
|
|
data.append(
|
|
|
|
self.meta_info(audio_id,
|
|
|
|
meta_info(audio_id,
|
|
|
|
float(duration), wav,
|
|
|
|
float(duration), wav,
|
|
|
|
int(start), int(stop), spk_id))
|
|
|
|
int(start), int(stop), spk_id))
|
|
|
|
return data
|
|
|
|
return data
|
|
|
|
|
|
|
|
|
|
|
|
def load_speaker_to_label(self):
|
|
|
|
def load_speaker_to_label(self):
|
|
|
|
|
|
|
|
if not self.spk_id2label_path:
|
|
|
|
|
|
|
|
logger.warning("No speaker id to label file")
|
|
|
|
|
|
|
|
return
|
|
|
|
|
|
|
|
spk_id2label = {}
|
|
|
|
with open(self.spk_id2label_path, 'r') as f:
|
|
|
|
with open(self.spk_id2label_path, 'r') as f:
|
|
|
|
for line in f.readlines():
|
|
|
|
for line in f.readlines():
|
|
|
|
spk_id, label = line.strip().split(' ')
|
|
|
|
spk_id, label = line.strip().split(' ')
|
|
|
|
self.spk_id2label[spk_id] = int(label)
|
|
|
|
spk_id2label[spk_id] = int(label)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
return spk_id2label
|
|
|
|
|
|
|
|
|
|
|
|
def convert_to_record(self, idx: int):
|
|
|
|
def convert_to_record(self, idx: int):
|
|
|
|
sample = self.data[idx]
|
|
|
|
sample = self.data[idx]
|
|
|
|
|
|
|
|
|
|
|
|
record = {}
|
|
|
|
record = {}
|
|
|
|
# To show all fields in a namedtuple: `type(sample)._fields`
|
|
|
|
# To show all fields in a namedtuple: `type(sample)._fields`
|
|
|
|
for field in type(sample)._fields:
|
|
|
|
for field in fields(sample):
|
|
|
|
record[field] = getattr(sample, field)
|
|
|
|
record[field.name] = getattr(sample, field.name)
|
|
|
|
|
|
|
|
|
|
|
|
waveform, sr = load_audio(record['wav'])
|
|
|
|
waveform, sr = load_audio(record['wav'])
|
|
|
|
|
|
|
|
|
|
|
|
# random select a chunk audio samples from the audio
|
|
|
|
# random select a chunk audio samples from the audio
|
|
|
|
if self.config.random_chunk:
|
|
|
|
if self.config and self.config.random_chunk:
|
|
|
|
num_wav_samples = waveform.shape[0]
|
|
|
|
num_wav_samples = waveform.shape[0]
|
|
|
|
num_chunk_samples = int(self.config.chunk_duration * sr)
|
|
|
|
num_chunk_samples = int(self.config.chunk_duration * sr)
|
|
|
|
start = random.randint(0, num_wav_samples - num_chunk_samples - 1)
|
|
|
|
start = random.randint(0, num_wav_samples - num_chunk_samples - 1)
|
|
|
|
@ -71,46 +98,9 @@ class VoxCelebDataset(Dataset):
|
|
|
|
# we only return the waveform as feat
|
|
|
|
# we only return the waveform as feat
|
|
|
|
waveform = waveform[start:stop]
|
|
|
|
waveform = waveform[start:stop]
|
|
|
|
record.update({'feat': waveform})
|
|
|
|
record.update({'feat': waveform})
|
|
|
|
record.update({'label': self.spk_id2label[record['spk_id']]})
|
|
|
|
if self.spk_id2label:
|
|
|
|
|
|
|
|
record.update({'label': self.spk_id2label[record['lab_id']]})
|
|
|
|
return record
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def __getitem__(self, idx):
|
|
|
|
|
|
|
|
return self.convert_to_record(idx)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def __len__(self):
|
|
|
|
|
|
|
|
return len(self.data)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class RIRSNoiseDataset(Dataset):
|
|
|
|
|
|
|
|
meta_info = collections.namedtuple('META_INFO', ('id', 'duration', 'wav'))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def __init__(self, csv_path):
|
|
|
|
|
|
|
|
super().__init__()
|
|
|
|
|
|
|
|
self.csv_path = csv_path
|
|
|
|
|
|
|
|
self.data = self.load_csv_data()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def load_csv_data(self):
|
|
|
|
|
|
|
|
data = []
|
|
|
|
|
|
|
|
with open(self.csv_path, 'r') as rf:
|
|
|
|
|
|
|
|
for line in rf.readlines()[1:]:
|
|
|
|
|
|
|
|
audio_id, duration, wav = line.strip().split(',')
|
|
|
|
|
|
|
|
data.append(self.meta_info(audio_id, float(duration), wav))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
random.shuffle(data)
|
|
|
|
|
|
|
|
return data
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def convert_to_record(self, idx: int):
|
|
|
|
|
|
|
|
sample = self.data[idx]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
record = {}
|
|
|
|
|
|
|
|
# To show all fields in a namedtuple: `type(sample)._fields`
|
|
|
|
|
|
|
|
for field in type(sample)._fields:
|
|
|
|
|
|
|
|
record[field] = getattr(sample, field)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
waveform, sr = load_audio(record['wav'])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
record.update({'feat': waveform})
|
|
|
|
|
|
|
|
return record
|
|
|
|
return record
|
|
|
|
|
|
|
|
|
|
|
|
def __getitem__(self, idx):
|
|
|
|
def __getitem__(self, idx):
|
|
|
|
|
xiongxinlei
3 years ago