PaddlePaddle版本：2.1.2
自定义的Dataset片段：
class AudioDataset(Dataset):
..............
defgetitem(self, idx):
p = self.files[idx]
filename = os.path.basename(p)
speaker = filename.split(sep='_', maxsplit=1)[0]
label = self.encoder.transform([speaker])[0]
mcep = np.load(p)
mcep = paddle.to_tensor(mcep, dtype='float32')
mcep = paddle.unsqueeze(mcep, 0)
print(paddle.to_tensor(speakers.index(speaker), dtype='int64'))
return mcep, paddle.to_tensor(speakers.index(speaker), dtype='int64'), paddle.to_tensor(label, dtype='float32')

测试：
def data_loader(datadir: str, batch_size=4, shuffle=True, num_workers=2):
dataset = AudioDataset(datadir)
loader = DataLoader(dataset, batch_size=batch_size, shuffle=shuffle, num_workers=num_workers)
return loader
dloader = data_loader(config.data_dir, batch_size=config.batch_size, num_workers=config.num_workers)
data_iter = iter(dloader)
x,index,label=next(data_iter)
print('Test')
print(index)

输出：
Tensor(shape=[1], dtype=int64, place=CPUPlace, stop_gradient=True,
[3])
Tensor(shape=[1], dtype=int64, place=CPUPlace, stop_gradient=True,
[2])
Test
Tensor(shape=[1, 1], dtype=int64, place=CPUPlace, stop_gradient=True,
3)
Tensor(shape=[1], dtype=int64, place=CPUPlace, stop_gradient=True,
[1])