jerky speech from doubletalk?

[Kerry Hoath]

This is not a processor speed issue exactly but a kernel design decision.
Most systems have what is called a system clock or timer routine or similar.
This routine handles scheduling and the time of day clock and other system functions.
It is similar to the hardware timer interrupt that o ccurrs under dos
at 18.2 times per second; (55 ms per tick) which keeps the clock r unning.
Under Linux this timer function is usually called 100 times a second.
this works fine on slower cpus and does not impact the system too much. It also allows
the time of day clock to stay rather accurate.
Processers have come a long way since Linus wrote his first code on a
386 SX/16 and many people want better multi-tasking and faster
responsiveness from the kernel.
One way to achieve this is to make the time of a jiffy shorter.
If the timeslice for multitasking is shorter then you can get more tasks to run in
the same time because the units of time are smaller.
New kernels have upped the hz value to 512 or 1024 on a 868-style
kernel. You can't do this on a 386 or 486 because there wouldn't be enough
time for code to execute with the jiffy t imer r unning so
quickly.
However since processers have got a lot faster; we have plenty of spare
cycles and the increased jiffy timing will actually improve system
response lower latency and help real-time projects.
You can tinker the timer values for your speech synth or turn off the hz
modifications in the kernel. Not sure how long these will remain
optional so it looks like speakup might need a bit of tweaking.
Do not fear; your shiny new machine will run Linux just fine.
Redhat and Mandrake just have a habbit of dropping ht elatest new
features on an unsuspecting public.
Anyone remember gcc 2.96 <grin>

Regards, Kerry.
On Wed, Aug 28, 2002 at 02:20:15PM -0400, Charles Crawford wrote:
> Huh?  I just ordered a Pentium 4 with 2.5 gigahertz processing from 
> pololinux. So am I going to be looking at a real speeech problem?  I use a 
> Dectalk express.
> 
> -- Chrlie.
>    At 10:05 AM 8/28/02 -0600, you wrote:
> >Hi Aaron,
> >
> >      You nailed the problem.  The i386 kernel would fix it, but there's an
> >easier way.  Speakup lets you adjust the timing parameters, so it's
> >possible to go into the /proc/speakup/<synth> directory and use echo or cp
> >to set the values for the 512Hz kernel.  I use settings similar to these.
> >jiffy_delta=25 and delay_time=100.
> >I use an Audapter, but I tried to adjust the values based on the litetalk
> >defaults.  The trigger_time value is radically different between the
> >Audapter and Litetalk.  I didn't change mine, but it's possible that you
> >might need to.
> >      It should be pointed out that we're messing with timers actually in
-- 
Kerry Hoath:  kerry at gotss.net kerry at gotss.eu.org or  kerry at gotss.spice.net.au
ICQ: 8226547 msn: kerry at gotss.net Yahoo: kerryhoath at yahoo.com.au