This would enable music, dialogue, sfx and background ambiences to start and stop independently and play simultaneously at runtime.

Ideally, some form of looping/ crossfading/ blocking/ ducking would be natively supported as well, though this could potentially be handled on the server or client side if the audio stream was able to be left open at great length.

Some form of this is already built into the Alexa stack as can be observed when Alexa interrupts music being played: In this case music doesn't stop but rather it ducks while TTS audio fires simultaneously.

Exposing this feature to the Skill community would allow for the same amount of real-time, dynamic audio mixing freedom as what exists as a standard feature on the majority of gaming platforms.