snr#

diffsptk.SNR#: alias of SignalToNoiseRatio

class diffsptk.SignalToNoiseRatio(frame_length: int | None = None, full: bool = False, reduction: str = 'mean', eps: float = 1e-08)[source]#

See this page for details.

Parameters:

frame_lengthint >= 1 or None: The frame length in samples, \(L\). If given, calculate the segmental SNR.
fullbool: If True, include the constant term in the SNR calculation.
reduction[‘none’, ‘mean’, ‘sum’]: The reduction type.
epsfloat >= 0: A small value to avoid NaN.

forward(s: Tensor, sn: Tensor) → Tensor[source]#

Calculate SNR.

Parameters:

sTensor [shape=(…, T)]: The signal.
snTensor [shape=(…, T)]: The signal with noise.

Returns:

outTensor [shape=(…,) or scalar]: The SNR.

Examples

>>> import diffsptk
>>> import torch
>>> snr = diffsptk.SignalToNoiseRatio(full=True)
>>> s = torch.tensor([0.1, -0.2, 0.3, -0.4, 0.5])
>>> n = s * 0.1
>>> y = snr(s, s + n)
>>> y
tensor(20.0000)

diffsptk.functional.snr(s: Tensor, sn: Tensor, frame_length: int | None = None, full: bool = False, reduction: str = 'mean', eps: float = 1e-08) → Tensor[source]#

Calculate SNR.

Parameters:

sTensor [shape=(…, T)]: The signal.
snTensor [shape=(…, T)]: The signal with noise.
frame_lengthint >= 1 or None: The frame length in samples, \(L\). If given, calculate the segmental SNR.
fullbool: If True, include the constant term in the SNR calculation.
reduction[‘none’, ‘mean’, ‘sum’]: The reduction type.
epsfloat >= 0: A small value to avoid NaN.

Returns:

outTensor [shape=(…,) or scalar]: The SNR.