How to measure latency and see how good your audio interface REALLY is!

Latency is the biggest problem in audio today. Yet once, it didn't even exist. But how can latency be measured and one audio interface truly compared against another?

Yes it is true. Latency - the delay between a signal entering an audio system and that same signal leaving - once didn't exist. That was in the days of analog, when the time taken for a signal to enter and exit a mixing console, recorder or signal processor was zero point zero. Literally.

But they had other problems - noise for instance, which is arguably even worse. (The argument would be between noise, which affects the technical quality of the recording but not an artist's performance, and latency, which affects the performance, but not the technical quality. Discuss!)

The biggest problem with latency is that it exists. The second biggest problem is how to measure it, because it is only by comparison that you can find out which system has the lowest latency. And it is only by comparison that developers will be compelled to achieve the lowest figures possible.

Latency is caused in the audio interface on the input side, in the software, by the computer, and once more in the audio interface on the output side.

All this makes it very difficult to compare measurements because there are so many possibilities. Any suggestions for a method of isolating the latency due to each of these causes individually would be welcome indeed!

But you can measure and compare latency for yourself. All you need to do is set up a 'latency party', similar to a LAN party where computer gamers bring their computers physically together, hook up a network and play.

The idea is to gather together several audio interfaces. This is the most significant item in the latency chain to measure for two reasons - a) you already know that a faster computer will have less latency, and b) you would almost certainly rather change your interface than change your recording software.

So here's what you do... first choose a computer to run the tests on. I would choose a fast, recent model, one that is 'clean' without any unnecessary software installed; one that can be restored easily once the tests are done.

Set up the system to record from a microphone onto one channel, and simultaneously record the output of that channel onto another channel. Be careful with the routing otherwise the test will be invalid.

The result will be two identical signals, one delayed with respect to the other. Zoom in and measure the delay in milliseconds directly, or take the measurement in samples and calculate the latency by dividing by the sampling rate.

To give an example, I measured a Digidesign Mbox and got a figure of 42 milliseconds - terrible, but I didn't expect miracles from a slowish PC.

In comparison, I measured the latency of a Pro Tools MixPlus PCI interface and achieved a figure of 1.8 milliseconds. Yes, one point eight milliseconds!

Clearly I am comparing a very inexpensive interface with a very expensive PCI interface with DSP. Similarly priced interfaces will give closer figures. But they will be different, and they will give you a very good guide as to which audio interface is the best one for your configuration.

(One small point - you might think that it would be enough to take an already recorded waveform on one track, then record it to another track through the interface. The software designers aren't that crazy though - the delay will be compensated for, otherwise as you build up a multitrack recording track by track, delays would accumulate throughout the whole process.)