Learning Time-Graph Frequency Representation for Monaural Speech Enhancement