比较与torchaudio.transforms.GriffinLim的差异

torchaudio.transforms.GriffinLim

class torchaudio.transforms.GriffinLim(n_fft: int = 400, n_iter: int = 32, win_length: Optional[int] = None, hop_length: Optional[int] = None,
                                       window_fn: Callable[[...], torch.Tensor] = <built-in method hann_window of type object>, power: float = 2.0,
                                       normalized: bool = False, wkwargs: Optional[dict] = None, momentum: float = 0.99,
                                       length: Optional[int] = None, rand_init: bool = True)

更多内容详见torchaudio.transforms.GriffinLim。

mindspore.dataset.audio.GriffinLim

class mindspore.dataset.audio.GriffinLim(n_fft=400, n_iter=32, win_length=None, hop_length=None,
                                         window_type=WindowType.HANN, power=2.0,
                                         momentum=0.99, length=None, rand_init=True)

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

差异对比

PyTorch：使用Griffin-Lim算法从线性幅度频谱图中计算信号波形。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：使用Griffin-Lim算法从线性幅度频谱图中计算信号波形。

分类	子类	PyTorch	MindSpore	差异
参数	参数1	n_fft	n_fft	-
	参数2	n_iter	n_iter	-
	参数3	win_length	win_length	-
	参数4	hop_length	hop_length	-
	参数5	window_fn	window_type	MindSpore仅支持5种窗函数
	参数6	power	power	-
	参数7	normalized	-	STFT后幅值规范化，MindSpore不支持
	参数8	wkwargs	-	自定义窗函数的入参，MindSpore不支持
	参数9	momentum	momentum	-
	参数10	length	length	-
	参数11	rand_init	rand_init	-

代码示例

import numpy as np

fake_input = np.ones((151, 36)).astype(np.float32)

# PyTorch
import torch
import torchaudio.transforms as T
torch.manual_seed(1)

transformer = T.GriffinLim(n_fft=300, n_iter=10, win_length=None, hop_length=None, window_fn=torch.hann_window, power=2, momentum=0.5)
torch_result = transformer(torch.from_numpy(fake_input))
print(torch_result)
# Out: tensor([-0.0800,  0.1134, -0.0888,  ..., -0.0610, -0.0206, -0.1800])

# MindSpore
import mindspore as ms
import mindspore.dataset.audio as audio
ms.dataset.config.set_seed(3)

transformer = audio.GriffinLim(n_fft=300, n_iter=10, win_length=None, hop_length=None, window_type=audio.WindowType.HANN, power=2, momentum=0.5)
ms_result = transformer(fake_input)
print(ms_result)
# Out: [-0.08666667  0.06763329 -0.03155987 ... -0.07218403 -0.01178891 -0.00664348]