比较与torchaudio.transforms.TimeMasking的差异
torchaudio.transforms.TimeMasking
class torchaudio.transforms.TimeMasking(time_mask_param: int, iid_masks: bool = False)
mindspore.dataset.audio.TimeMasking
class mindspore.dataset.audio.TimeMasking(iid_masks=False, time_mask_param=0, mask_start=0, mask_value=0.0)
差异对比
PyTorch:给音频波形施加时域掩码。
MindSpore:给音频波形施加时域掩码。不支持变化的mask_value
取值。
分类 |
子类 |
PyTorch |
MindSpore |
差异 |
---|---|---|---|---|
参数 |
参数1 |
time_mask_param |
time_mask_param |
- |
参数2 |
iid_masks |
iid_masks |
- |
|
参数3 |
- |
mask_start |
添加掩码的起始位置 |
|
参数4 |
- |
mask_value |
指定填充掩码值,MindSpore计算时无法再更改 |
代码示例
import numpy as np
fake_wav = np.array([[[0.17274511, 0.85174704, 0.07162686, -0.45436913],
[-1.0271876, 0.33526883, 1.7413973, 0.12313101]]]).astype(np.float32)
# PyTorch
import torch
import torchaudio.transforms as T
torch.manual_seed(1)
transformer = T.TimeMasking(time_mask_param=2, iid_masks=True)
torch_result = transformer(torch.from_numpy(fake_wav), mask_value=0.0)
print(torch_result)
# Out: tensor([[[ 0.0000, 0.8517, 0.0716, -0.4544],
# [ 0.0000, 0.3353, 1.7414, 0.1231]]])
# MindSpore
import mindspore as ms
import mindspore.dataset.audio as audio
ms.dataset.config.set_seed(2)
transformer = audio.TimeMasking(time_mask_param=2, iid_masks=True, mask_start=0, mask_value=0.0)
ms_result = transformer(fake_wav)
print(ms_result)
# Out: [[[ 0. 0.85174704 0.07162686 -0.45436913]
# [ 0. 0.33526883 1.7413973 0.12313101]]]