Build the demo

[dvdsk]

from the repo:

Currently Rodio is by far the most widely used audio engine. Unfortunately it seems limited by what we knew 9 years ago about audio development in Rust. Since then new crates have been build to overcome that. We hope to demonstrate one crate to be best suited to all use cases and then build support behind it. For that we are building a demo. It will take the form of a game environment since game audio is seen as the most challenging use case. The goal is to have users feel how the game responds differently when switching audio engine.

Currently two crates are participating in this demo: Rodio and Firewheel though everyone is very welcome. The demo will be limited to functionality that is properly implemented by all crates.

Demo

• audio environment playing back many sounds at the same time.
• sounds and effects are influenced by user input to make it easy to judge latency.
• the demo uses audio libraries in the same way as a game.
• pre-defined events instead of user input to keep the demo repeatable.
• no graphics to keep the demo code simple.
• sounds are all at the same sample-rate.
• implementations pre-resample before loading.

We know what we want from the demo, now it just needs to be build. I see 2 stages

• build the game part
• integrate the audio engines
  I am not not be the best person for the first part, as I have zero ECS experience. I will do the integration of Rodio as audio engine. @BillyDM would you be okay with integrating Firewheel?

Then we still need someone to build the game part. My hope is that one of us can take this on. If that turns out not to be the case we can ask for help in the rust audio and bevy discord. If we do ask for help that should be combined with a small announcement of this effort. Any announcement needs to wait till all text is final which it not yet is.

[alice-i-cecile]

I'm happy to provide review on the game part, but I don't have the time to tackle it on my own.

[dvdsk]

I'll be taking this on in the next week(s).

We will need many different sounds, than we can really stress the audio engines while having something fun to play with.

Examples of sounds I would like to have: crickets, all kinds of birds, leaves rustling, rain, thunders, animals (deer, wolves?), walking (multiple surfaces?).

Does anyone know a good place to find (free) audio assets like those?

[alice-i-cecile]

https://sonniss.com/gameaudiogdc19/ and similar for other years are an excellent choice here. We want high quality sounds, but aren't too picky about what is used.

[dvdsk]

Its been three weeks so let me give a status update: I've been making good progress in the time I've had, though I did get distracted as bevy's Transform introduced me to quaternions. That can be quite a mathematical rabithole. Which was fun.

I hope to submit a minimal game PR this week and then add some animals moving around (making noise) after that. To that end I've found a very nice (mathematical) description of bird flocks.

Regarding the game, its going to get minimal visuals (animal = sphere or cube). Since that makes debugging and implementing the sound system far simpler.

[CorvusPrudens]

Ah, looks like I missed this effort! I've put something together myself, although it may not have the full variety of sounds we'd be looking for in a true stress test.

[dvdsk]

Ah, looks like I missed this effort! I've put something together myself, although it may not have the full variety of sounds we'd be looking for in a true stress test.

Looks very nice. Your quite far ahead of me. It looks super nice! I would love to help build it out more! One thing I was working on is procedural bird floks. The idea was to increase flok size till one of the engines fails (stats artifacting/lagging).

[dvdsk]

So what I've wrote is, at its core, a modified version of bevy's 3d example. The demo by CorvusPrudens is really lovely I would like to stay close to it.

• It also works well for testing audio latency. You press each time to go to the next scene. A sound is supposed to play immediately. If later in the demo the sound is delayed you surely will feel it.
• Not having any 3d renderer makes it far easier for others to read through.

So how do we add a ton more sounds without destroying the demo's story?

[dvdsk]

One idea, when they cross the creek we could have a bunch of randomized frogs/insects and maybe the bird floks? But can we throw in ~200 sounds without it destroying the demo's ..... serenity?

[CorvusPrudens]

So how do we add a ton more sounds

Yeah maybe the idea wasn't the best for this sort of thing! It can be quite tricky to create a natural-sounding soundscape without just crowding everything out.

But I suspect you're right. Procedural things like simulating a bunch of frogs or a flock of birds should give us the ability to throw tons of sounds at the engines. The chimes are a very basic example of that, where each time they're triggered, we play a bunch of different samples in succession with progressively lower volume.

Note that, as-is, the failure modes of the two engines would be pretty different. With Firewheel as it's set up, it'll just drop incoming samples if there are no available samplers. Nothing exciting happens. rodio, on the other hand, will keep adding more until it starts under-running.

Firewheel can be made to add more samplers if it runs out, but it would be kind of annoying to do for the demo. bevy_seedling has this capability built-in, but there's a lot of infrastructure in place to make it convenient. So I guess we'll have to figure out something there.

[CorvusPrudens]

We could just pre-allocate a ton of samplers. I did a quick test with 256 total, and while that's maybe a little excessive for normal usage, it would help us demonstrate playing hundreds of samples at a time pretty easily.

[dvdsk]

We could just pre-allocate a ton of samplers. I did a quick test with 256 total, and while that's maybe a little excessive for normal usage, it would help us demonstrate playing hundreds of samples at a time pretty easily.

Lets do that 👍, seems like a good stress test. For rodio I would like to see if I can use the mixer instead of opening multiple streams. That should perform better when throwing hunderds of sounds at it.

[dvdsk]

Procedural things like simulating a bunch of frogs or a flock of birds should give us the ability to throw tons of sounds at the engines.

And we can make the number a parameter to the engine. Might help finding the breaking point for each engine.

unrelated:

The timings are taken from the Core Audio client

That is really quite nice. Might make for a good automated benchmark too.

[BillyDM]

Just a heads up, I'm making a fairly significant change to Firewheel that should make it much easier for node developers. I'll let you know when it is done!

[BillyDM]

Alright, the new changes are done!