cutcutcodec.core.signal.psd.welch

cutcutcodec.core.signal.psd.welch(signal: Tensor, freq_resol: Real = None) → Tensor[source]

Estimate the power spectral density (PSD) with the Welch method.

Parameters

signaltorch.Tensor: The stationary signal on witch we evaluate the PSD. The tensor can be batched, so the shape is (…, n).
freq_resolfloat, default=10: The norlised frequency resolution in Hz, assuming a sample rate of 1. It is the lowest denoised frequency as well. Higher it is, better is the resolution but noiser it is.

Returns

psdtorch.Tensor: An estimation of the power spectral density, of shape (…, m).

Examples

>>> import torch
>>> from cutcutcodec.core.signal.psd import welch
>>> signal = torch.randn((32, 2, 768000))
>>> psd = welch(signal)
>>>
>>> # freq = torch.fft.rfftfreq(2*psd.shape[-1]-1, 1/48000)
>>> # import matplotlib.pyplot as plt
>>> # _ = plt.plot(freq, psd[0].T)
>>> # plt.show()
>>>