Hi Dylan, please use plain text in your messages, HTML makes it hard to quote your text... Your idea of having fewer wakes makes sense. Just technically I think you are confusing latency with frame size. if you want to use 10ms frames for speech processing, you will have a 20ms latency, be that with ALSA or PulseAudio. Also I think you've hit an issue with PulseAudio's inner details. The idea is that there's a server side buffer that has the same length than the ALSA ring buffer. This makes sense for low-power audio, so that you can wake-up at the last moment and quickly fill the ring buffer. For low-latency, this might not be such a good idea, since it entails many useless wakes and the client does have the data available. This behavior is enforced in pulsecore/protocol-native.c, it may be possible to patch this code to reduce the server side buffer to zero (or minreq). This might be a good point to bring to Lennart, if he still remembers what he wrote a couple of years ago. Lennart are you still with us? -Pierre
