Speaking vs. singing shouldn't matter. As in a play, you have to be able to handle both.
Sensitivity is the issue. In certain cases the engineer may have to ride the fader on the spoken word vs. that which is sung due to the power of the singing voice. This should not be affected by the gain setting, which should remain constant but rather by the fader.
Also, a limiter inserted either into the individual channel, a sub-group of vocals or the main outputs of the mix itself easily remedies transients or uneven levels of output.
Typically, when mixing a singer live, the effects (delay, reverb) are patched through one or two channels in the board (or stereo channels in some cases) so they can be easily muted when the speaker is not singing. Of course, these effects can also be routed through the aux sends and returns but there is no muting option on these. In this case you could use the bypass button on the effects unit if one exists. Fader control gives you more flexibility and range than a rotary pot and you have the ability to easily bring levels up and down with the effects.
One troublesome method of mixing I have witnessed on more than one occasion is the "all fader, minimum gain" technique. I do not know where this originated but it is a disturbing practice. Some "engineers" will use a full throttled fader on a vocal channel with hardly any gain level. Not only do you limit the headroom of your amplifiers using this technique but if you are mixing monitors through the FOH you are limiting the output through the aux(monitor) send. I have had a good laugh many times watching guys using this method struggle trying to get any appreciable level to the monitor mixes.
Of course, always use pre-fader aux sends for FOH monitor mixes as well as effects to keep levels even or unaffected while increasing fader level. A post-fader send will increase volume of monitor level and effects while bringing up the channel's level. This will almost always result in feedback in more ways than one.