I worked on this Unity project as part of Studio13, a Canadian game audio company led by Joel Walsh. It's a top-down city-builder RTS, with quite a detailed visual style. The project had some notable technical challenges to solve.
Listener placement
Placing the listener in a top-down 3D game isn't completely straightforward. You could just have it follow the camera directly, but that feels off because whatever is going on just below and behind you is going to be more audible than what's going on in front of you. The visual focus is on the ground in front of you, and the sonic focus should follow suit.

Example: Listener is placed at the y-level of the camera, but x and z of the ground level where the camera is pointing.

Example: Listener is placed between the camera and the ground level, by a configurable amount. This is the solution we ended up with, with some tweaks.
In Tunic (2022), the camera view is isometric. They had a similar approach to listener placement. They also used a super neat of applying audio attenuation based on screen space. This was explained in a GDC talk by Kevin Regamey.
Clusters of sound
At any moment in the game, the player might be overlooking hundreds of NPCs. Some are chopping wood, some are soldiers battling, and so on. If we'd implement audio on each and every one, the performance would likely have been abysmal. This would also be hard to control in terms of volume and overall spread of sound.
To solve this issue, Joel, Mark and I landed on an approach where each unit is treated as a point in a cluster. Our "cluster rig" is a sphere that moves with the camera. Any time a unit enters the sphere, we track its position. For example, if 8 battling soldiers are closer than 20 meters to each other, they are treated as a singular group, with one looping sound playing instead of all of them playing their individual sounds. That way, we save a bit of audio overhead while being able to design a loop that captures the unique sound of an army.
There's a bit more nuance to it, with regards to what state units or buildings are in and how it moves with the camera. The player can zoom all the way out to reveal the world map, at which point the cluster rig moves up into the air - effectively muting the audio of whatever is happening on the ground. We are also checking the state of buildings or units to differentiate buildings on fire, for example.

An example of how a single type of building might be laid out in a cluster.

Some clusters might be more complex. This makes placement of our single loop audio source more tricky.
Dynamic music
To support the gameplay, we needed a dynamic music system. Having experience writing such systems before, I was tasked with creating it. The game has certain events that we tap into to alter the music. By default, the player hears soft, idle exploration music. We made sure to add randomized lengths of silence between each track. At some point, buildings and people in the game can become sick. Whenever this happens, or a building catches fire, the music crossfades into a more unnerving and unsettling soundtrack. If a battle breaks out between players, dramatic battle music plays.
We made sure to handle prioritization, in case any of these things were to happen simultaneously. If a battle happens before a sickness breaks out, we should continue to hear the battle music since that's more important for the player to handle. We tested this by implementing controls and playback visualization directly in the inspector, to be able to easily debug the system.

The music system inspector.
Each of the three states (Calm, Challenge and Struggle) have their own playlist of 4-5 tracks. This, combined with the silent periods in-between them makes it so players won't grow tired of it too quickly.


