Text this: A Comparative Study of Deep Audio Models for Spectrogram- and Waveform-Based SingFake Detection