Rust-native audio transcoding

[marceline-cramer]

Closes #38.

Regrettably, there are no pure-Rust ogg Vorbis encoding crates right now. However, I've added the next best thing: vorbis_rs. It has active maintainers and uses the latest ogg libraries. The other Rust ogg libraries are unmaintained and use older versions of libogg and libvorbis with known security vulnerabilities.

I also caught an issue where sound graph files were using the string literal "SOUND_GRAPH_EXTENSION" instead of the constant value SOUND_GRAPH_EXTENSION, so I've patched that.

Also, I haven't implemented the copying of tags from the input file to the output file. The reason is that vorbis_rs expects tag values to be strings but symphonia tag values can be lots of different types. I'm not sure what to do about this, so I've left tags empty in the output file for now.

[chaosprint]

Just played with the physics examples using a wav file and got this error:

[2023-04-17T07:13:58Z INFO  symphonia_core::probe] found the format marker [52, 49, 46, 46] @ 0+2 bytes.
[2023-04-17T07:13:58Z ERROR ambient_build] In pipeline pipeline.json/1, at file ping.wav
    
    Caused by:
        0: Failed to create audio decoder
        1: unsupported feature: core (codec):unsupported codec

I will now try to fix it on top of your code.

It seems that the hint is not used at all inside symphonia:

symphonia::default::get_probe().format()

The real cause is CodecType 264 is not in the CodecRegistry codecs HashMap

---

When I enable all default features in symphonia, I got:

[2023-04-17T08:35:07Z ERROR ambient_build] In pipeline pipeline.json/1, at file bonk.wav
    
    Caused by:
        Expected 2 channels in audio block, got 3

which is very strange...

---

well, you need to use "pcm" feature rather than "wav" for symphonia

Even so, I still got an error. symphonia somehow probes wav as an mp3...

[2023-04-17T08:55:51Z INFO  ambient] Building Physics
ext "wav"
[2023-04-17T08:55:51Z INFO  symphonia_core::probe] found the format marker [ff, fd] @ 0+51 bytes.
[2023-04-17T08:55:51Z WARN  symphonia_bundle_mp3::demuxer] invalid mpeg audio header
[2023-04-17T08:55:51Z WARN  symphonia_bundle_mp3::demuxer] skipping junk at 6542 bytes
[2023-04-17T08:55:51Z WARN  symphonia_bundle_mp3::demuxer] skipping junk at 18532 bytes
[2023-04-17T08:55:51Z WARN  symphonia_bundle_mp3::demuxer] skipping junk at 20470 bytes
[2023-04-17T08:55:51Z WARN  symphonia_bundle_mp3::demuxer] skipping junk at 22824 bytes

---

Right, three features have to be enabled: ["wav", "pcm", "mp3"] (not very intuitive but I will pr them later...)

I got the same error when all features are enabled:

When I enable all default features in `symphonia`, I got:

[2023-04-17T08:35:07Z ERROR ambient_build] In pipeline pipeline.json/1, at file bonk.wav

Caused by:
    Expected 2 channels in audio block, got 3

[chaosprint]

Hi,

It looks that Symphonia still has some work to do. So we have to wait or switch to other solutions.

We can use minimp3 and hound instead. Are you interested to do the switch?

[marceline-cramer]

I would be interested in doing the switch, but symphonia is very close to being fully working. Would the minimp3 or hound crates provide any features that symphonia wouldn't? Symphonia supports other audio codecs like FLAC and AAC, and adding support to those with symphonia is as simple as enabling the feature flag and matching the audio asset pipeline with the corresponding file extensions.

It seems like now that .wav files are being processed correctly, the problem is that the bits in the channel flags are counted to set the channel count in the Vorbis encoder. This doesn't produce the correct number of channels in all cases. This can be fixed if I use a more comprehensive method of querying which channel bits are set instead, and I'll be sure to cover more testcases with the test input files.

[chaosprint]

For extra featues, probably not. But I do have minimp3 and hound working on my Mac while for symphonia, I cannot run its official examples. Have you tried to run the symphonia examples?

[marceline-cramer]

Yeah, I've ran symphonia-play from the latest master branch and it works fine on every audio file I've given it so far. For extra context I'm on Artix Linux. Symphonia also has passing CI on Mac. Do you know what's keeping you from running the examples?

[marceline-cramer]

I made a mistake with the channel count; counting the bits WAS the correct way to do that, but I confused the .bits() function (which returns the numeric bitfield) with the .count() function (which actually counts the bits).

After enabling the pcm Symphonia feature and counting the bits correctly, the stereo .wav files I've tested now get converted to ogg with no issue.

[chaosprint]

I made a mistake with the channel count; counting the bits WAS the correct way to do that, but I confused the .bits() function (which returns the numeric bitfield) with the .count() function (which actually counts the bits).

After enabling the pcm Symphonia feature and counting the bits correctly, the stereo .wav files I've tested now get converted to ogg with no issue.

Cool, I will test this ASAP

[philpax]

Sounds good to me - the log hiding is done at the top-level here:

Ambient/app/src/main.rs

Lines 28 to 40 in 40eb6eb

	LevelFilter::Warn,
	&[
	"ambient_build",
	"ambient_gpu",
	"ambient_model",
	"ambient_physics",
	"ambient_std",
	"cranelift_codegen",
	"naga",
	"tracing",
	"wgpu_core",
	"wgpu_hal",
	],

We can just add a commit to this PR to do that, and merge (Marceline won't be online until later) - are you OK to do that?

[chaosprint]

The latest version works perfectly on my Mac. During the conversion, some unnecessary logs are printed, but I guess it's harder to remove as this comes from the symphonia itself. We can work on the conversion mechanism later: #320

[2023-04-19T08:21:42Z INFO  symphonia_core::probe] found the format marker [49, 44, 33] @ 0+2 bytes.
[2023-04-19T08:21:42Z INFO  symphonia_core::probe] found the format marker [ff, fb] @ 1246+2 bytes.
[2023-04-19T08:21:42Z INFO  symphonia_bundle_mp3::demuxer] using xing header for duration```

[chaosprint]

Sounds good to me - the log hiding is done at the top-level here:

Ambient/app/src/main.rs

Lines 28 to 40 in 40eb6eb

	LevelFilter::Warn,
	&[
	"ambient_build",
	"ambient_gpu",
	"ambient_model",
	"ambient_physics",
	"ambient_std",
	"cranelift_codegen",
	"naga",
	"tracing",
	"wgpu_core",
	"wgpu_hal",
	],

We can just add a commit to this PR to do that, and merge (Marceline won't be online until later) - are you OK to do that?

Yes