ChannelVolume misbehaving with non power of 2 number of channels

[DivineGod]

I seem to have stumbled into a strange issue where adding more output channels to a ChannelVolume will make it mess up the output.

My setup is that I have a custom source which is basically an extended version of Spatial with a self.input which is a ChannelVolume if I create an audio interface with 2 or 4 channels and configure the ChannelVolume to have 2 or 4 channels the output seems fine. But if I add more than 4 the output which was meant to go to channel 1 goes to channel 5 and channel 2 goes to channel 6.

Shown here is the loopback interface where the volume is shown for the 6 channel version:

Here is the version with 4 channels which shows the correct behaviour:

The only thing changed is the number of channels in the output configuration which impacts ChannelVolume getting either 4 or 6 channels. I have confirmed that the set_volume call does get called with the correct index and volume levels as per my spatial algorithm in both cases.

Edited to add:

Upon further inspection it seems that any number of channels that are not powers of two say 6 or 10 will exhibit this behaviour while power of two number of channels look like they behave perfectly okay

[DivineGod]

Still trying to make a reduce repro case. Currently my code seem to have the wrong data buffer after dynamic mixer_rx iterator is run.

I tried copying my setup in a test case in the dynamic mixer with a ChannelVolume with 6 channels and a SampleBuffer source with a few samples but it did not convert into the expected error. So I’ll keep digging into this.

[DivineGod]

Found this comment (https://github.com/RustAudio/rodio/pull/320#issuecomment-703079059) and that made me think it might be related as the frame length might not be evenly divisible with a non-pow 2 number of channels.

I’ll investigate further and see what I find

[DivineGod]

The above didn't seem to give me anything on an input source that takes from a device, but from a file there might be something even worse going on. I've got crackling and weirdness going on with an ogg file using rodio::Decoder into my dbap sink (spatial sink with more than 2 channels basically) which has a channel volume inside

[DivineGod]

The above didn't seem to give me anything on an input source that takes from a device, but from a file there might be something even worse going on. I've got crackling and weirdness going on with an ogg file using rodio::Decoder into my dbap sink (spatial sink with more than 2 channels basically) which has a channel volume inside

Crackling due to codec being in 44100Hz and output in 48000Hz. Really not useful. Hopefully it makes sense to someone and there's a way to fix this?

[DivineGod]

I managed to recreate the issue by modifying one of the examples:

use std::io::BufReader;

use rodio::source::ChannelVolume;

fn main() {
    let (_stream, handle) = rodio::OutputStream::try_default().unwrap();
    let sink = rodio::Sink::try_new(&handle).unwrap();

    let file = std::fs::File::open("assets/music.ogg").unwrap();
    let input = rodio::Decoder::new(BufReader::new(file)).unwrap();
    let chan_vol = ChannelVolume::new(input, vec![0.5, 0.5, 0.0, 0.0, 0.0, 0.0]);
    sink.append(chan_vol);

    sink.sleep_until_end();
}

I've used Loopback to create a 6 channel output device and macOS Audio MIDI Setup utility to make sure the sample rate is correct and that the device is selected as the default output device. This is done so I didn't have to mess with selecting the right config in the code.

Attached screenshot of the Loopback interface showing output on channels 5 and 6 even though as seen by the code it's meant to be channels 1 and 2

[DivineGod]

I tried using a SineWave source in my example and had the output device set to 44.1kHz to match the music.ogg source file, as it seemed that resampling didn't work.

Using the SineWave I can confirm that mapping to different sampling rates breaks things in more ways when there's a ChannelVolume in the mix, but if there isn't then it seems to be fine. This is still with 6 channels as opposed to say 2 or 4 which seem to work fine.

[DivineGod]

Instead of using a sink to control play I tried with handle.play_raw() and it seems to behave as I would expect, no issues with channels getting mixed up.

[DivineGod]

Looking at what is being output from the mixer_rx in new_output_stream_with_format in src/stream.rs the data buffer is initially being filled with data where the first 12 samples are the same value followed by 24 samples of 0.0 then repeated until the end of the buffer at 3072 samples (512 * 6)

The next time through, the data ends up with being 4 samples of 0.0 then 2 samples of data, which is then repeated until the end of the buffer.

First off it seems weird that the data is 12 + 24 samples the first time through the mixer and secondly that it then sort of corrects itself but not quite after that. I do notice the first bit of weird data as an odd click and can see the channels activate in loopback.

[DivineGod]

ChannelCountConverter is wrong for the first iteration is most likely what's happening with that.

[DivineGod]

I think I've got a fix. It's mostly to do with Empty having 1 channel it messes up everything. It should have 0 channels, because it doesn't actually do anything and then there should be special cases handling such a case, because in that case you want to have the output channel number used.

[DivineGod]

I spoke too soon. The fixes worked for the case of SineWave but an ogg file is still messed up and now the channels are 3+4 instead of 5+6. An only if the sample_rate is correct for the output and input.

[DivineGod]

I've identified a number of issue stemming from the use of the Emtpy source when initially populating the queue in Sink it would cause the various conversions to fail so I've implemented some fixes that seem to fix that.

[DivineGod]

I've marked #493 as ready for review. It deals with the immediate issue of channels mapping wrongly. The Sample Rate conversion issue I'll investigate in a separate PR