LWN: Comments on "Experimental applications of GStreamer"

Audio location

smurf — Fri, 13 Nov 2015 19:53:21 +0000

Latency may be variable, but once you start a recording it, well, records like it's supposed to. Thus, if the sound of all speakers are on a single recording and you evaluate the differences in timing between speakers instead of the sounds' absolute arrival times, there's no problem.

Audio location

nye — Thu, 12 Nov 2015 15:31:15 +0000

I don't see how that help in any way with the problem that the latency is wildly variable?

Audio location

smurf — Sun, 08 Nov 2015 15:16:01 +0000

You didn't understand my idea; let me rephrase.

As I understand it, the senders don't have delay issues, they're all on the receiving end.

Thus, instead of pinging one speaker at a time and measuring the absolute time the pulse takes to arrive at the listener, you ping all of them at the same time, record that sound, and use digital processing to figure out the delays between pings. (Thus different frequencies, otherwise you can't tell the pings apart.) This way, delays on the receiver side won't matter, it's all a single sound recorded once by a single microphone.

Audio location

oldtomas — Thu, 05 Nov 2015 11:24:28 +0000

> Why would the absolute timing of audio at the receiver matter?

It wouldn't if it were constant. But (from the article):

"the Android-introduced delays varied between 30 to 100ms"

I read that as having an uncertainty of roughly 70ms. Of course, if you could develop a solid statistical model of this uncertainty you could try to pick up the signal from the noise, but that sounds like a tall order compared to just cancelling out a (per-device) constant...

Experimental applications of GStreamer - tiled streaming

sciurus — Sat, 31 Oct 2015 19:29:36 +0000

Yes, AWS offers GPU instances with GRID K520 and Tesla M2050 cards from NVIDIA.

Audio location

smurf — Thu, 29 Oct 2015 18:50:52 +0000

Why would the absolute timing of audio at the receiver matter?
Send a ping to each speaker, at the exact same time but at different frequencies, then measure relative timing at the receiving end.
With known (fixed) speaker locations, calculating the receiver's position is a non-problem these days.

Experimental applications of GStreamer - tiled streaming

jnareb — Thu, 29 Oct 2015 17:04:28 +0000

| It would be better to use GPUs for the stitching and tiling, he said, but so far no cloud provider offers such a service.

Actually they are cloud providers that offer GPGPU computing access. For example QwikLab, which offers paid CUDA and OpenACC courses (first lab in series is free) runs on Amazon cloud.

Experimental applications of GStreamer

paulj — Thu, 29 Oct 2015 09:52:46 +0000

Any old arena in Amsterdam, or Amsterdam Arena?