比较与torchaudio.transforms.Spectrogram的差异

torchaudio.transforms.Spectrogram

class torchaudio.transforms.Spectrogram(n_fft: int = 400, win_length: Optional[int] = None, hop_length: Optional[int] = None,
                                        pad: int = 0, window_fn: Callable[[...], torch.Tensor] = <built-in method hann_window of type object>,
                                        power: Optional[float] = 2.0, normalized: bool = False, wkwargs: Optional[dict] = None,
                                        center: bool = True, pad_mode: str = 'reflect', onesided: bool = True)

更多内容详见torchaudio.transforms.Spectrogram。

mindspore.dataset.audio.Spectrogram

class mindspore.dataset.audio.Spectrogram(n_fft=400, win_length=None, hop_length=None,
                                          pad=0, window=WindowType.HANN,
                                          power=2.0, normalized=False,
                                          center=True, pad_mode=BorderType.REFLECT, onesided=True)

更多内容详见mindspore.dataset.audio.Spectrogram。

差异对比

PyTorch：从音频信号创建其频谱。支持自定义窗函数或对窗函数传入不同的配置参数。支持对STFT结果进行幅值规范化。

MindSpore：从音频信号创建其频谱。

分类	子类	PyTorch	MindSpore	差异
参数	参数1	n_fft	n_fft	-
	参数2	win_length	win_length	-
	参数3	hop_length	hop_length	-
	参数4	pad	pad	-
	参数5	window_fn	window	MindSpore仅支持5种窗函数
	参数6	power	power	-
	参数7	normalized	normalized	-
	参数8	wkwargs	-	自定义窗函数的入参，MindSpore不支持
	参数9	center	center	-
	参数10	pad_mode	pad_mode	-
	参数11	onesided	onesided	-

代码示例

import numpy as np

fake_input = np.array([[[1, 1, 2, 2, 3, 3, 4]]]).astype(np.float32)

# PyTorch
import torch
import torchaudio.transforms as T

transformer = T.Spectrogram(n_fft=8, window_fn=torch.hamming_window)
torch_result = transformer(torch.from_numpy(fake_input))
print(torch_result)
# Out: tensor([[[[3.5874e+01, 1.3237e+02],
#                [1.8943e+00, 3.2839e+01],
#                [8.4640e-01, 2.1553e-01],
#                [2.0643e-02, 2.4623e-01],
#                [6.5697e-01, 1.2876e+00]]]])

# MindSpore
import mindspore.dataset.audio as audio

transformer = audio.Spectrogram(n_fft=8, window=audio.WindowType.HAMMING)
ms_result = transformer(fake_input)
print(ms_result)
# Out: [[[[3.5873653e+01 1.3237122e+02]
#         [1.8942689e+00 3.2838711e+01]
#         [8.4640014e-01 2.1552797e-01]
#         [2.0642618e-02 2.4623220e-01]
#         [6.5697211e-01 1.2876146e+00]]]]