Le 10/05/2021 à 09:12, Mario Lang a écrit :
Didier Spaier <did...@slint.fr> writes:
Le 09/05/2021 à 23:09, Samuel Thibault a écrit :
Didier Spaier, le dim. 09 mai 2021 23:02:49 +0200, a ecrit:
Still, this could be an issue for audiophiles, but they can use jack.
Speech-based screen reading also requires very low latency to get
snappy
feedback.
OK, then does anyone knows how to measure the latency increase?
Latency in RT audio is mostly driven by buffer size.
These buffer sizes vary greatly, depending on the software and hardware
in use. JACK with a good soundcard can indeed go down to
a buffer size of 64 frames or even lower. That makes for a fixed
latency of around 2ms, depending on the sample rate.
On the ohter end of the spectrum, carelessly written audio programs
might go up to buffer sizes of up to 2048. Now latecy is more around
50ms, which is starting to get noticeable.
However, I suspect there are several latencies adding up in the case of
speech synthesis. It would indeed be worthwhile to investigate if some
of these can be reduced. However, I think this would need some
automated testing.
To come back to your question, can you specify a little more clearly
which increase you are refering to?
I am speaking about latency that could occur if the sound is mixed
first by pulseaudio then by dmix from alsa, following Aura's message:
http://brltty.app/pipermail/brltty/2021-May/018429.html
of which I paste the content below:
On 2021-05-09 at 22:28 +0200, Didier Spaier <didier at slint.fr> wrote:
> ### In Slint, we want to share audio resources between speech apps that
> ### rely on alsa and other apps that rely on pulseaudio.
> load-module module-alsa-sink device=dmix
> load-module module-alsa-source device=dsnoop
[--]
> 1. I have been told that such a setting is not recommended. But nobody
> so far
> has been able to tell me why, and I have received zero complaint
> from users
> about it so far, as they get speech and braille in both console and
> graphical
> environments and can easily switch.
The reason is latency. If the audio is mixed in software both on pulseaudio
and alsa levels, the latency is unnecessarily high, and the processing will
waste resources.
<end of pasted content>
Cheers,
Didier
_______________________________________________
This message was sent via the BRLTTY mailing list.
To post a message, send an e-mail to: BRLTTY@brltty.app
For general information, go to: http://brltty.app/mailman/listinfo/brltty
