Caution

You're reading an old version of this documentation. If you want up-to-date information, please have a look at 0.10.2.

librosa.util.softmask

librosa.util.softmask(X, X_ref, *, power=1, split_zeros=False)[source]

Robustly compute a soft-mask operation.

M = X**power / (X**power + X_ref**power)

Parameters:

Xnp.ndarray

The (non-negative) input array corresponding to the positive mask elements

X_refnp.ndarray

The (non-negative) array of reference or background elements. Must have the same shape as X.

powernumber > 0 or np.inf

If finite, returns the soft mask computed in a numerically stable way

If infinite, returns a hard (binary) mask equivalent to X > X_ref. Note: for hard masks, ties are always broken in favor of X_ref (mask=0).

split_zerosbool

If True, entries where X and X_ref are both small (close to 0) will receive mask values of 0.5.

Otherwise, the mask is set to 0 for these entries.

Returns:

masknp.ndarray, shape=X.shape: The output mask array

Raises:

ParameterError

If X and X_ref have different shapes.

If X or X_ref are negative anywhere

If power <= 0

Examples

>>> X = 2 * np.ones((3, 3))
>>> X_ref = np.vander(np.arange(3.0))
>>> X
array([[ 2.,  2.,  2.],
       [ 2.,  2.,  2.],
       [ 2.,  2.,  2.]])
>>> X_ref
array([[ 0.,  0.,  1.],
       [ 1.,  1.,  1.],
       [ 4.,  2.,  1.]])
>>> librosa.util.softmask(X, X_ref, power=1)
array([[ 1.   ,  1.   ,  0.667],
       [ 0.667,  0.667,  0.667],
       [ 0.333,  0.5  ,  0.667]])
>>> librosa.util.softmask(X_ref, X, power=1)
array([[ 0.   ,  0.   ,  0.333],
       [ 0.333,  0.333,  0.333],
       [ 0.667,  0.5  ,  0.333]])
>>> librosa.util.softmask(X, X_ref, power=2)
array([[ 1. ,  1. ,  0.8],
       [ 0.8,  0.8,  0.8],
       [ 0.2,  0.5,  0.8]])
>>> librosa.util.softmask(X, X_ref, power=4)
array([[ 1.   ,  1.   ,  0.941],
       [ 0.941,  0.941,  0.941],
       [ 0.059,  0.5  ,  0.941]])
>>> librosa.util.softmask(X, X_ref, power=100)
array([[  1.000e+00,   1.000e+00,   1.000e+00],
       [  1.000e+00,   1.000e+00,   1.000e+00],
       [  7.889e-31,   5.000e-01,   1.000e+00]])
>>> librosa.util.softmask(X, X_ref, power=np.inf)
array([[ True,  True,  True],
       [ True,  True,  True],
       [False, False,  True]], dtype=bool)