Michael Prokop <mika at grml.org> writes: > * C.M. Brannon <cmbrannon at cox.net> wrote: > >> There's a very easy fix for this: >> renice 3 `ps -e |grep speechd-up |head -1 |cut -d' ' -f1` > [...] > > Are we talking about the same grml version? > grml 1.0 automatically does a 'nice -n -20 speechd-up' when invoking > swspeak. Does not that fix your issue? Hi Mika, I'm using the latest and greatest, version 1.0. I have better success when speechd-up has a positive (low) priority, rather than a negative one. I think this is because a low priority process makes fewer reads to /dev/softsynth, so it is more likely to read words, rather than single characters. You can actually view this with a packet capture tool, reading incoming messages on port 6560 (used by speech-dispatcher). When speechd-up runs with priority <= 0, I see a speak message generated and sent to speech-dispatcher for every character in a word, but when it runs with priority > 0, it usually sends a speak message to dispatcher containing a whole word or line of text. I really don't have an explanation for this, especially considering that other people are not encountering the same behavior that I am! I think the solution lies in modifying the speechd-up sources to use a different buffering strategy, rather than recompiling kernels and changing process priorities... -- Chris
