Responding here to a PM request from the OP-
How to best approach it will depend on the material as well as your target listening situation, personal preferences, and how much effort you want to put into it.
The overarching thing is keeping the reduction in dynamic range unnoticeable to listeners.. unless they were to compare it to the original full range version. When a relatively large reduction of range is required, it helps to combine a few different strategies, which in combination achieves the desired reduction transparently, even though any one approach on its own would be too much and become audibly obvious if called upon to make the full reduction all by itself. Layering 2, 3 or more different dynamic range reduction strategies allows each one of them to be sufficiently conservative that none of them stick out as obvious, yet collectively achieves the dynamic range reduction you want. You can use this multi-stage dynamics strategy for any live recording, but it may not be necessary and requires additional work.
Start by reducing errant peaks that are significantly higher than everything else. You can identify those visually. Look for those that really stick out. If just a few you can manually redraw them. Or set a limiter that isn't working all the time but is just catching the few excessive peaks. It should have a high threshold, very fast attack (look ahead) and relatively fast release. Its only job is to tame the highest peaks. This step is technically reducing the overall dynamic range, but not really in a perceptual way. Its more of a housekeeping thing that makes the following steps more productive.
Maybe make some manual volume envelope type adjustments. Stuff like manually bringing up the level of quieter sections, bringing down the level of the audience reaction if its louder than the performance, bringing down the excessively loud parts. I moved this here because it is more straight-forward to do by drawing volume changes on the computer screen than is the appropriate setting of compressor settings. Draw in fast or slow fade ups/downs at the start and end of broad plateau sections to make the changes inaudible. This technique is most akin to automating the manual volume knob adjustments you would otherwise make while listening.
Maybe apply parallel compression to bring otherwise barely audible low level details up. This is like automatically turning up just the quieter sounds more than the loud sounds. Its one of the more powerful ways of reducing dynamic range transparently because it very effectively reduces range without squashing the energetic parts. Check out the thread here at TS on parallel compression for how to apply it. Essentially you squash a copy of the recording pretty strongly with compression using a low threshold, and mix that in with the non-compressed copy. The non-compressed copy retains its dynamic energy whenever it gets loud. Bring up the level of the compressed copy until the low level parts are heard clearly, but not so much that the noise floor or audience chatter is increased too much. It can help to EQ the squashed copy with a loudness-like curve that complements quieter listening levels. If the recording is in need of noise reduction, this will be more difficult as it will make the noise or any noise-reduction efforts more audible, but can also make appropriate noise-reduction efforts especially worthwhile.
Maybe apply standard top down compression. This in my experience is harder to get sounding really transparent. It's like automatically turning down the volume when things get loud. It is tricky because it is reducing level in a more easily perceptual way when things get energetic, then increasing level again as soon as that energy dissipates somewhat. This is harder to do without killing the performance energy and/or hearing the compression actively working.
Maybe apply a mastering type limiter. That essentially brings the entire energy level up a touch more and smooths everything out, and is the last thing done in combination with normalizing / setting the final peak output level. Plugin versions of these limiters have gotten a lot more transparent and easier to use over the years. If not, adjust final output level manually so that the highest peaks are just a bit under 0dBfs.
Generally, this multi-pronged approaching works well because it achieves the needed dynamic range reduction via a few different ways: Over very short, medium and long timescales, and by doing so manually in addition to automatically. Whatever techniques you end up using, listen closely at each step and don't apply so much of any one that it ever becomes obvious. One of the more difficult but most productive things is the listening art of learning how to take yourself out of control-adjustment-mindset mode (knowing specifically what you are listening to, for, and adjusting) and switching over to a sort of third-party perceptual listening mode where you are attempting to not be specifically aware of what you were just doing, but rather are hearing it as someone else would. That's one of the more difficult things to do honestly without self-deception, but really benefits all post-recording mixdown decisions.
Topic: Smoothing out a singer/songwriter performance (Read 1290 times)