Re: GNOME 2 Sound Architecture and APIs?

[Alan Cox <alan redhat com>]

> The big issue for accessibility, at least, is latency.  Presumably the 
> sound event framework allows sounds to be played on a near-real-time 
> basis, provided they have been pre-loaded.

This reminds me there is another requirement for the new audio API. We must
pass a textual or visual description of the event and its importance so that
deaf users can select visual descriptions for these events.

> text-to-speech will require of the sound subsystem:
> latency <= 100 ms
> ability to stop sound emission
> ability to readily mix sound requests to play concurrently

Realistically the hardware limits are something like

	Cannot assume anything but 16bit 48KHz fixed frequency
	Cannot assume multiple channels
	Cannot assume audio can be killed in less than .2 seconds
			(about 5 seconds worst case on a few very old cards)

The only way we will hit <=100ms latency nicely is to use the XSync extension
and hook X up to the audio layer so you can schedule display and audio events
at a given point in the audio time sequence.

Alan