In addition to the line level to tape head level attenuation, you would need to do the opposite of the tape player equalization so that the signal resembles the tape pulses output.
Chances are pretty good that the waveform from the signal generator will be cleaner and wider bandwidth (i.e. more "square") vs. the signal from the playback tape head (more "rounded" corners of the pulse waveform). IOW, the signal generator signal could have nice sharp corners of the pulses because it hasn't been recorded to and replayed from tape.
So, depending on what is the point of the experiment, you could simulate the roll-off of the record->tape->playback path with a simple high-frequency roll-off. The higher the frequency or the wider the head gap, or the lower the speed of the tape, the more high frequency roll-off you could expect.
OTOH you could just leave it "clean" if it is not important to the overall exercise.
If you are working with a regular waveform (like a sync clock), then a simple waveform (perhaps a square wave with the appropriate duty-cycle) out of a common signal generator would be good. But if it is a more complex waveform (like SMPTE timecode, etc.) then many modern signal generators (even low-budget models) will support generating arbitrary waveforms as appropriate for your experiment.