Hello,

Following PR #390, while I was able reproduce the issue fixed by the PR, I still had issues with the sound output after the change.

So I downloaded the source code for rodio and cpal to try to find the cause. I've tested various methods of setting up the audio output and was able to narrow down the issue to Sink::try_new(). I've updated the test code to retry repeatedly to make the issue more prominent. On my setup (Arch Linux x86_64; Ryzen 5600X; B550 chipset) with the sound mostly going to the wrong channel about 70% of the time, and to the correct channel 30% of time.

extern crate rodio;

use rodio::source::Source;
use rodio::cpal;
use rodio::cpal::traits::HostTrait;
use rodio::cpal::traits::DeviceTrait;

fn main() {
    let device = cpal::default_host().default_output_device().expect("failed to find output device");
    let config = device.default_output_config().unwrap();

    let sample_rate = config.sample_rate().0;

    let (stream, stream_handle) = rodio::OutputStream::try_from_device(&device).unwrap();

    // left = sine wave; right = silence.
    let mut lt_ch = sine_wave(sample_rate);
    let mut rt_ch = vec![0.0; lt_ch.len()];

    let data: Vec<f32> = lt_ch.iter().zip(rt_ch.iter()).flat_map(|(&a, &b)| vec![a, b]).collect();

    for i in 0..50 {
        let buffer = rodio::buffer::SamplesBuffer::new(2, sample_rate, data.clone());
        let sink = rodio::Sink::try_new(&stream_handle).unwrap();
        sink.append(buffer.repeat_infinite());
        std::thread::sleep_ms(200);
    }
}

fn sine_wave(sample_rate: u32) -> Vec<f32> {
    let OSC_FREQ  : f32   = 440.0;
    let OSC_OMEGA : f32   = std::f32::consts::TAU * OSC_FREQ as f32;
    let OSC_PERIOD: usize = (sample_rate as f32 / OSC_FREQ) as usize;
    let OSC_GAIN  : f32   = 0.5;

    (0 .. OSC_PERIOD).map(|i| f32::sin(OSC_OMEGA * (i as f32 / sample_rate as f32)) * OSC_GAIN).collect()
}

While debugging the rodio, I found that SourcesQueueOutput does not guarantee that samples are output with the correct frame boundary. Furthermore, it creates a hardcoded 441 samples of silence as a buffer when there are no sources. I think this causes sources to be output with the wrong frame boundary when they are appended to the sink.

For the stereo output setup I was testing with, changing the zero buffer to an even counted size seems to mitigate the problem. Though for a proper fix, I think frame boundary guarantees needs to be factored into the API.

Left and right channels seem to be swapped #286

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions