This example demonstrates how to implement a simplified version of the Audio Random Container using nested generators. Specifically, it covers the following:
- A simple scheduler that picks children either sequentially or randomly and triggers them at a fixed interval.
- Polyphony with oldest-voice replacement; the realtime side maintains an array of voice slots and recycles the oldest as new trigger events arrive.
- A random level offset applied to each new trigger.
- A clean split between the control part and the realtime part, with bidirectional events used for lifetime coordination.
The project targets Unity 6.5 and is built on the Scriptable audio pipeline APIs.
Open the project in Unity 6.5, open Assets/Scenes/Random Container.unity, and enter Play Mode.
The scene wires three preset Random Containers in Assets/RandomContainers/ to their own AudioSources. Each button triggers one of them, demonstrating a different pattern:
Forest.asset: many short clips chosen at random with level jitter; a classic ambience layer.Machine_Gun.asset: a single clip retriggered at tight intervals; shows voice overlap and decay.Count_Master.asset: nested containers. Its children are themselves Random Containers, demonstrating that anIAudioGeneratorparent does not care what kind of generator its children are.
A Random Container is a ScriptableObject (RandomContainerAsset in Assets/Scripts/RandomContainer/RandomContainer.cs) that you drop into the Generator field of an AudioSource. The inspector exposes five fields:
| Field | Meaning |
|---|---|
children |
The pool of generators to choose from. Any IAudioGenerator works: AudioClips, other Random Containers, or your own custom generators. |
mode |
Sequential cycles through children; Random picks children uniformly. |
maxPlayCount |
How many total triggers to produce before going silent. |
interval |
Interval between triggers in seconds. |
randomizeLevel |
± volume range in dB applied to each triggered child. |
The implementation is located in Assets/Scripts/RandomContainer/ and is split across six files. The split follows the control vs realtime model described in the manual.
| File | Role |
|---|---|
RandomContainer.cs |
The ScriptableObject entry point. Implements IAudioGenerator.CreateInstance to wire up the control / realtime pair. |
RandomContainer/Control.cs |
Main-thread side; owns child references, drives the scheduler, creates and destroys child generators in response to triggers and realtime-side eviction requests. |
RandomContainer/Realtime.cs |
Audio-thread side; mixes the active child voices into the output buffer. Burst-compiled. |
RandomContainer/Scheduler.cs |
A small accumulator that emits a TriggerEvent every interval seconds. |
RandomContainer/Events.cs |
The message types exchanged over the pipe. |
RandomContainer/Defaults.cs |
Default inspector values and the Mode enum. |
The two halves communicate through a single bidirectional pipe carrying InstanceEvent messages:
- Control -> Realtime: "Here is a new child instance, start mixing it into a voice slot."
- Realtime -> Control: "I am done with this instance (it either finished by itself or it was replaced), please destroy it."
Voices live in a fixed-size pool of slots. When a new instance arrives and all slots are full, the oldest one is kicked out and reused for the new instance.
The control side stores its children behind GCHandles in a NativeArray. See the comment block at the top of Control.cs for why that pattern is necessary when bridging managed and unmanaged memory.
This is meant as a teaching example. A few things are intentionally limited:
- Fixed output format: The container ignores the host's suggested speaker mode and sample rate, leaving any conversion to the host.
- Hardcoded polyphony: No voice-stealing heuristics beyond "pick oldest".
- Not sample-accurate: Trigger events are truncated to buffer boundaries rather than placed at exact sample positions.
The source files are commented. Reading them top-to-bottom alongside the Generators and Concepts pages is the fastest way to understand the code.