Please everyone do the following very easy test?

Hi all, I recently uncovered a strange phenomenon. I am wondering if this is an issue that is endemic to my machine, or if, in fact, it’s not really an issue at all because everyone’s program/computer behaves this way.

So the request is this: Do a latency test. In case you are wondering how, please see here:

But I want you to tweak the test. This time, instead of just recording the test track once, do it multiple times, each time creating a new track. (IMPORTANT: Remember to mute each test track so it’s not re-recording those tests: that original click you create should be the only one not muted.) Record about five test tracks, one right after the other. They don’t have to be long. Maybe even just five seconds each.

NOW, here is what I am curious about. Zoom in. WAY IN. Forget about how the test tracks sync up with the original track. I’m actually not interested in that. What I am interested in this: Are all of your test tracks EXACTLY synced with each other? In theory, they SHOULD be. After all, you are recording the same exact thing from the same exact spot without changing anything.

What I discovered today, which has been driving me nuts, is that my tracks are actually NOT sync’d with one another. They vary from a few thousandths to a few hundredths of a second. I am curious to know if, in reality, everyone’s programs/machines behave like this, or if this is truly an issue I need to address.

Thanks so much in advance. Oh, and also, when you post your results, please let me know if you’re using a mixing board, or if you’re just plugging directly into the soundcard. Any kind of color around what your set up is would be awesome. If you’re curious what inspired this whole thing, see my other thread here:

Here’s mine.
MacBook Pro 10.5.8 (current version is 10.8.3)
112GB Solid State Drive with linux-based operating system and 24GB free.

Default click track.

I went through the whole production system. Microphone, mixer, headphones, computer Line-In to Line-Out. I didn’t bother to set Recording Latency. I just made four separate latency tests from track one by stuffing the microphone into the headphone muff and turning everything up. I get a repeatability of, what, maybe one or two milliseconds, if that?

Please note I’m not using the Windows Operating System and I’m not using a spinning oxide hard drive. Because of the SSD, my machines open and close software much faster than “normal” machines. Even they have measurable errors, but they’re really tiny. Also please note that about a fifth of the drive is open and free. It’s not smashed full with software and data.

Picture 4.png

This is what you’re hearing. Ignore the machine on the right. Koz

Thank you, Koz! Given that you still get a small variance (especially when you consider all of what you told me), I am thinking that what I am getting might just be standard. Of course, this is based on a sample size of 1.

Would love to see/hear about others’ tests.

I’m on Linux, which offers choices of different sound servers.

The default sound system on this machine is PulseAudio, which is a general purpose sound system that supports multiple applications and multiple sound cards (even networked sound cards).

I also have 2 sound cards - an internal “on board” sound card and an external Behringer UCA 202 USB.
The UCA 202 is connected to a mixing desk.

With PulseAudio:

Internal sound card:
The required latency correction is about 50 ms more than the buffer size and will vary by up to +/- 10 ms

UCA 202
The required latency correction is about 70 ms more than the buffer size and will vary by up to +/- 10 ms

With Jack

The other sound system that I use is “Jack Audio System”. This is a high performance, low latency sound system (like ASIO on steroids).
The “absolute” latency of Jack is configured through its own settings. Because this is just a 5 year old cheap laptop PC I have Jack set up with relatively high (for Jack) latency of 23 ms (on a fast system this can be safely set to less than 10 ms).

I’ve only tested with Jack with the USB device (this is my usual set-up when recording).
The Audacity “Audio to buffer” is not used, so the “Latency correction” is an absolute figure for the round trip from my computer to the mixing desk and back. This is set to -38 ms. With these settings, each new recording is 2 samples early. Each track has identical timing.

Screenshot with Jack Audio System. The top track is the generated click track. The tracks below are the recorded tracks:


Fascinating! And fantastic. So with your other non-Jack configurations, you will get a latency variation up to +/-10ms! This really answers my question, or I should say, further supports the the theory that even when you set the latency correction at a static rate, it’s still going to differ from time to time.

That’s incredible that the Jack system returns IDENTICAL results, though. I’m pretty jealous!

The wonders of open source :slight_smile:

Yes, but usually within a couple of milliseconds. +/- 10 ms is the maximum (well actually about 8.5 ms).In absolute terms 10 ms is not very much but could sound slightly out.
Get 3 good drummers to hit their drums “at the same time” and they will almost certainly be a millisecond or two out from each other.

To get an idea of how close the timing needs to be so as not to sound wrong, I’ve written a little bit of code that can be run in Audacity to create an irregular click track:
If you run the above code to two tracks at the same time, with a “variance” setting of 5 ms, then each track could be +/- 5 ms, so the worst case would be 10 ms difference between beats in each track.