A way to reduce memory load is to stream the audio directly from optical disc, hard drive or flash memory. Streaming is great for large audio files that would otherwise take up large amounts of memory in the SoundBank; however, as it needs to load on triggering, a small amount of latency might be induced. As such, streaming is not ideal for sounds that rely on precise timing such as lip-synced dialogue, but is well adapted for sounds where timing is generally not critical such as ambience, and sometimes music.

Streaming is very similar to loading a SoundBank. When a SoundBank is loaded, the content is streamed to memory by the CPU. When streaming an audio file, a small amount of CPU is likewise used to fetch the streamed file from the hard drive or disc, and—depending on the settings you specify—the streamed data will either be flushed out or kept in memory.

Your streaming settings are configured on the Audio Object in your Wwise Project. Let’s demonstrate streaming on the Quest 4 Blacksmith dialogue Sound Voice Object.

	Sound Voice object is a type generally used for dialogue, specifically designed to handle multiple languages. Sound Voice object is not to be confused with a Voice.

By simply enabling Stream, on generation of the SoundBanks the sound will be stored in a WEM file next to the SoundBanks to be streamed in-game.

	A WEM (Wwise Encoded Media) file is a Wwise proprietary format archived in the compiled game.

The following options are available to streamed files:

The music, dialogue, and ambience in WAG are streamed, which saves approximately 57 MB of runtime memory. Within this, dialogue takes up approximately 37 MB, which underlines the importance of using the Non-cachable option when possible, to remove cache left over in memory.