A Generalized Subspace Approach for Multichannel Speech Enhancement Using Machine Learning-Based Speech Presence Probability Estimation