Complexscaleinvariantsignalnoiseratio

complexscaleinvariantsignalnoiseratio

Metric ComplexScaleInvariantSignalNoiseRatio from torchmetrics (torchmetrics.audio.ComplexScaleInvariantSignalNoiseRatio)

When to invoke this skill

The user has predictions + ground truth and asks to evaluate with ComplexScaleInvariantSignalNoiseRatio, or mentions torchmetrics.audio.ComplexScaleInvariantSignalNoiseRatio directly, or wants the standard torchmetrics implementation.

Reference signature

from torchmetrics.audio import ComplexScaleInvariantSignalNoiseRatio

# ComplexScaleInvariantSignalNoiseRatio(zero_mean: bool = False, **kwargs: Any) -> None

Library docstring

Calculate `Complex scale-invariant signal-to-noise ratio`_ (C-SI-SNR) metric for evaluating quality of audio.

As input to `forward` and `update` the metric accepts the following input

- ``preds`` (:class:`~torch.Tensor`): real float tensor with shape ``(...,frequency,time,2)`` or complex float
  tensor with shape ``(..., frequency,time)``

- ``target`` (:class:`~torch.Tensor`): real float tensor with shape ``(...,frequency,time,2)`` or complex float
  tensor with shape ``(..., frequency,time)``

As output of `forward` and `compute` the metric returns the following output

- ``c_si_snr`` (:class:`~torch.Tensor`): float scalar tensor with average C-SI-SNR value over samples

Args:
    zero_mean: if to zero mean target and preds or not
    kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.

Raises:
    ValueError:
        If ``zero_mean`` is not an bool
    TypeError:
        If ``preds`` is not the shape (..., frequency, time, 2) (after being converted to real if it is complex).
        If ``preds`` and ``target`` does not have the same shape.

Example:
    >>> from torch import randn
    >>> from torchmetrics.audio import ComplexScaleInvariantSignalNoiseRatio
    >>> preds = randn((1,257,100,2))
    >>> target = randn((1,257,100,2))
    >>> c_si_snr = ComplexScaleInvariantSignalNoiseRatio()
    >>> c_si_snr(preds, target)
    tensor(-38.8832)

Quick recipe

import torchmetrics.audio as _m
score = _m.ComplexScaleInvariantSignalNoiseRatio(y_true, y_pred)

Don'ts

• Don't reimplement when the library version handles edge cases (NaN, ties, empty inputs) better than a hand-rolled formula.
• Always check the library version's argument order — sklearn is (y_true, y_pred) while torchmetrics is (preds, target).