A non-procedural approach might be easier: collect loops of baseline murmurs / elevated cheering / peak ecstatic crowd and vary the volume of each based on what

A non-procedural approach might be easier: collect loops of baseline murmurs / elevated cheering / peak ecstatic crowd and vary the volume of each based on what is happening on screen. If you have variants of each "energy level" you can also smoothly fade between them, queuing the next one randomly as the current version runs out. This way the wrap-around point should not be that noticeable. Something similar is done for music that varies according to gameplay intensity: https://en.wikipedia.org/wiki/Adaptive_music https://en.wikipedia.org/wiki/Adaptive_music