Differences with torchaudio.transforms.Resample

torchaudio.transforms.Resample

class torchaudio.transforms.Resample(orig_freq: int = 16000, new_freq: int = 16000, resampling_method: str = 'sinc_interpolation')

For more information, see torchaudio.transforms.Resample.

mindspore.dataset.audio.Resample

class mindspore.dataset.audio.Resample(orig_freq=16000, new_freq=16000, resample_method=ResampleMethod.SINC_INTERPOLATION,
                                       lowpass_filter_width=6, rolloff=0.99, beta=None)

For more information, see mindspore.dataset.audio.Resample.

Differences

PyTorch: Resample a signal from one frequency to another.

MindSpore: Resample a signal from one frequency to another. Extra filter option is supported.

Categories	Subcategories	PyTorch	MindSpore	Difference
Parameter	Parameter1	orig_freq	orig_freq	-
	Parameter2	new_freq	new_freq	-
	Parameter3	resampling_method	resample_method	-
	Parameter4	-	lowpass_filter_width	Sharpness of the filter
	Parameter5	-	rolloff	The roll-off frequency of the filter
	Parameter6	-	beta	The shape parameter used for kaiser window

Code Exampleimport numpy as np

fake_input = np.array([[[[-0.2197528, 0.3821656]]],
                      [[[0.57418776, 0.46741104]]],
                      [[[0.76986176, -0.5793846]]]]).astype(np.float32)

# PyTorch
import torch
import torchaudio.transforms as T

transformer = T.Resample(orig_freq=16000, new_freq=24000)
torch_result = transformer(torch.from_numpy(fake_input))
print(torch_result)
# Out: tensor([[[[-0.2140,  0.2226,  0.3510]]],
#              [[[ 0.5728,  0.6145,  0.2789]]],
#              [[[ 0.7568, -0.1601, -0.6101]]]])

# MindSpore
import mindspore.dataset.audio as audio

transformer = audio.Resample(orig_freq=16000, new_freq=24000)
ms_result = transformer(fake_input)
print(ms_result)
# Out: [[[[-0.21398525  0.22255361  0.35099414]]]
#       [[[ 0.5728122   0.614469    0.2788692 ]]]
#       [[[ 0.75675076 -0.16008556 -0.61005235]]]]