Recording - April 2011 - 40

Latency: it’s a fact of life in computer recording. We have to deal with it, but to do that, we have to understand it....

SOME FACTS AND FICTION ABOUT LATENCY
By Mike Rivers
To paraphrase author-humorist Mark Twain’s comment about the weather, everyone talks about latency, but nobody does anything about it. Since most of today’s recording musicians are using some sort of computer-based multitrack DAW, “latency” is the frequent topic of tutorials and on-line discussions. Everyone knows that low latency is good, but what constitutes “low latency,” when is it important, and when isn’t it? What it is Latency is simply a time delay. Audio delay often serves a useful function, for example with echo or flanging effects or time-aligning loudspeakers. When delay becomes a nuisance, though, we call it “latency.” In a DAW, latency can involve both audio and MIDI signals, or, when playing a soft synth, a combination of the two. While latency can appear at many places in a signal chain, the most cussed-at in a DAW system is in the monitor path between the microphone and headphones. With a few diversions, that’s what we’ll address here. Early DAW adopters generally migrated from an analog recording setup which usually included a mixer. The mixer served as the hub of the system, routing mics or instruments to both the sound card inputs and headphones or monitor speakers. An analog signal path has essentially no delay—play a sound and you hear it immediately. Today’s computer-based DAWs include a software mixer which, like its analog counterpart, can also provide a headphone feed for tracking or overdubbing. While this approach simplifies setup and eliminates a piece of hardware, getting the signal through the DAW system takes some time. The time between when you sing a note and when you hear yourself in the phones is what we call monitoring latency. So what can we do to reduce DAW monitor latency? The DAW signal path To best attack latency, it’s helpful to understand the various sources of delay, what you can change, and what you can’t. Figure 1 shows a block diagram of the signal flow from mic to headphones in a basic DAW. The analog-to-digital and digital-to-analog converters each contribute a fixed delay of around 1.5 ms at 44.1 kHz sample rate (less at higher sample rates). The driver is the interface between hardware and software, passing data between the converters, operating system, and application program. Since the computer has other tasks in addition to handling your audio, input and output data are buffered to assure a steady flow in and out of the system. The input buffer acts as a “holding tank,” allowing data entering at a constant rate to fill the buffer. The buffer releases the data when the processor is ready to use it. Similarly, the output buffer receives data from the processor in spurts and sends it out in a steady stream. Audio number crunching is quite fast, though plug-ins and virtual instruments take additional processing time and therefore require larger buffers. Buffers must be large enough so that they never overflow, but the larger the buffer, the greater the delay. Buffer size is usually expressed as a number of samples. Buffer size divided by the sample rate, plus the converter latency, closely approximates system latency. How much latency is too much (or too little)? It depends. When there’s no need for input monitoring through the DAW, for example when recording a band playing on stage, or when mixing, latency is usually of no concern. Latency matters when you must listen to the output of the D/A converter in order to hear yourself. With too little buffering, the audio will stutter. Too much and latency will haunt you. Most people aren’t uncomfortable if mic-to-headphone delay is less than ten milliseconds. Since sound travels through air at about one foot per millisecond, a ten millisecond delay is similar to playing your guitar ten feet away from the amplifier. Anyone can do that, but 30 milliseconds starts sounding like an echo. On the other end, a very short delay (1–3 ms) can cause comb filtering, making your own voice sound odd in your headphones, though it will sound fine in playback. So what can you do? The simple answer is to make the buffers smaller. It could be as simple as checking a “Use low latency monitoring” check box; usually it’s more involved than that. Since I’m a Windows user, these tips are Windows-centric, but the principles apply to any
40
RECORDING APRIL 2011

B E T T E R L AT E N T T H A N N E V E R —

FIG. 1

Recording - April 2011

Table of Contents for the Digital Edition of Recording - April 2011