The only reason I even joined the discussion in this thread are the listed 5ms vs. 25ms audio latency numbers (assuming that they are credible of course). They make no sense. OK, we are adding software mixing, so we need some extra buffering in the pipeline. But why the latency is getting five times worse and not just twice? Is it because audio mixing in userspace is broken by design? Wouldn't it be natural to do the mixing in the hard irq handler, which is responsible for swapping dma buffers or something like this?




Reply With Quote
