Getting lowest latency sound?

Susan Cragin · Post by **Susan Cragin** » Mon Nov 24, 2008 8:27 am

I have been trying to get lowest-latency sound (with highest fidelity) to use with Dragon NaturallySpeaking.

I have Jaunty and the latest RT kernel, which I know has problems for many applications but works fine to run DNS. (It will not, however, install the program nor train it.)

I set up real-time audio access as follows:
sudo su -c 'echo @audio - rtprio 99 >> /etc/security/limits.conf'
sudo su -c 'echo @audio - nice -20 >> /etc/security/limits.conf'
sudo su -c 'echo @audio - memlock 1500000 >> /etc/security/limits.conf'

So as I understand it, rtprio is the correct #, nice is the lowest possible, and I memlocked half my RAM.

In winecfg, I have audio set to 48000 default sample rate.

I have an internal hda-intel soundcard (bad but better since alsa upgraded to 1.0.18). I also have a SoundBlaster X-Fi that doesn't work yet, but I'm waiting with my fingers crossed, since SoundBlaster open-sourced some of the code this month.

Any fairly simple suggestions?
How about a chrt command?

Thanks. Susan

Susan Cragin · Post by **Susan Cragin** » Mon Nov 24, 2008 8:42 pm

Hi Susan,

You should try using windows applications that use ASIO and use the Wine ASIO
driver, WINEASIO.

Regards,

-- Peter

Do you mean that my program, Dragon NaturallySpeaking, has to depend on ASIO when it runs in Windows? Dragon is really the only choice in speech recognition software.

Peter L Jones · Post by **Peter L Jones** » Tue Nov 25, 2008 11:18 am

Hi Susan,

You should try using windows applications that use ASIO and use the Wine ASIO
driver, WINEASIO.

Regards,

-- Peter

Susan Cragin wrote:

I have been trying to get lowest-latency sound (with highest fidelity) to use with Dragon NaturallySpeaking.

I have Jaunty and the latest RT kernel, which I know has problems for many applications but works fine to run DNS. (It will not, however, install the program nor train it.)

I set up real-time audio access as follows:
sudo su -c 'echo @audio - rtprio 99 >> /etc/security/limits.conf'
sudo su -c 'echo @audio - nice -20 >> /etc/security/limits.conf'
sudo su -c 'echo @audio - memlock 1500000 >> /etc/security/limits.conf'

So as I understand it, rtprio is the correct #, nice is the lowest possible, and I memlocked half my RAM.

In winecfg, I have audio set to 48000 default sample rate.

I have an internal hda-intel soundcard (bad but better since alsa upgraded to 1.0.18). I also have a SoundBlaster X-Fi that doesn't work yet, but I'm waiting with my fingers crossed, since SoundBlaster open-sourced some of the code this month.

Any fairly simple suggestions?
How about a chrt command?

Thanks. Susan

Peter L Jones · Post by **Peter L Jones** » Tue Nov 25, 2008 11:19 am

Susan,

I don't know the requirements for Dragon Naturally Speaking under Windows. The
application documentation should supply these.

I'm not that familiar with it but WINE now also supports direct JACK output.
You could also try that.

(I've not got a running Linux system with audio support to hand to check,
unfortunately -- I'm mid-rebuild.)

Regards,

-- Peter

Susan Cragin wrote:

Hi Susan,

You should try using windows applications that use ASIO and use the Wine
ASIO driver, WINEASIO.

Regards,

-- Peter
Do you mean that my program, Dragon NaturallySpeaking, has to depend on ASIO
when it runs in Windows? Dragon is really the only choice in speech
recognition software.