32d73ccb0d80db3727775fb9928f5f151ddaaab5,nt/speech_enhancement/mask_estimation.py,,simple_ideal_soft_mask,#,44

Before Change

    power = np.einsum("...dk,...dk->...k", X.conjugate(), X)
    mask = (power / np.sum(power, axis=power.ndim-1, keepdims=True)).real

    return mask


def quantile_mask(observations, quantile_fraction=0.98, quantile_weight=0.999):
     Calculate softened mask according to lorenz function criterion.

After Change

    if not tuple_output:
        return np.squeeze(mask, axis=feature_dim)
    else:
        sizes = np.cumsum([o.shape[source_dim] for o in input])
        output = np.split(mask, sizes[:-1], axis=source_dim)

        for i in range(len(output)):
            if output[i].shape[source_dim] is 1:
                output[i] = np.squeeze(output[i])
                // ToDo: Determine, why the commented code is not working
                // output[i] = np.squeeze(output[i], axis=(source_dim,feature_dim))
            else:
                output[i] = np.squeeze(output[i], axis=feature_dim)

        return output


def quantile_mask(observations, quantile_fraction=0.98, quantile_weight=0.999):

Instances