How far apart temporally can two sounds be heard while still sounding like they are playing at the same time?

Question

I am trying to synchronize music played on different computers. I've been collecting quantitative information to try to identify sources of asynchrony.

I would like to know how far apart in time two sounds need to be in order for people to think that they are playing at the same time. In other words, how imprecise can my synchronization be in order to sound like synchronization?

This obviously depends on the sort of music that is being played, so I think I'm looking for the lowest number, which I think would correspond to especially fast and crisp music.

October 6 update

Let's try asking it a different way. I've been looking for all of the possible sources of variable lag and logging them. Since the music still sounds off, at least one of the following must be true

1. There's a bug in my code.
2. I'm missing a source of lag.
3. A difference of one centisecond in the playing of two sounds is enough for people to hear them as different.

One centisecond is one beat at 6000 bpm, or about one sixty-fourth of a beat at 94 bpm.

October 10 update

Hmm I could also test this by offsetting a few sounds by a few centiseconds in some sound program, encoding them and then playing them.

And people seem to think that I'm streaming music or that I'm asking about sources of latency. I'm not streaming music, but I'll keep in mind everyone's thoughts on where the latency may be coming from.

Answer 1

I dug a up thesis on the topic, On the contribution of temporal and spatial cues to the perceptual organization of concurrent sounds (link). From scanning section 1.2 Temporal cues, there are at least three factors involved, namely attack shape, volume, and onset time. There's probably a frequency content dependency as well (i.e subbass/kick combos can have more temporal slop than cymbal/snare combos). However, as a rule of thumb (based on the referenced research by Schutte), it sounds like 50ms is a good approximate maximum offset for fast attacks. I didn't read the spatialization parts, but it sounds like the direction of the sound can be a factor as well.

Answer 2

I misunderstood your question so this is the edited version:

This is a big challenge specially in big monsters shows with only one sound source. They use a special hardware to split correctly the sounds with the correctly delay to make sounds "in sync" everywhere.

With more than one source you will need some net connection between the machines. In my opinion, better avoid wi-fi and use cables to get the high speed and security and no big delay between machines. Using sharing control you can control both in one machine and put machines to play "in sync".

You have to consider the sound velocity (330 meters/second), your area and the arrangement design of your sounds boxes to make a sound map ensuring every point will recept sound "equaly" with very little acceptable delay and with no interferences between boxes (too high volume generating a echo effect). Without a specialized sound enginner you can lose some time in a trial-and-error method but probably you can make it. I suggest hiring one, if your budget easy allow for this.

If you want

massively surround sound that nobody would be able to shut down,

you probably will look for rent some oil power generators too.

Good Luck with your speakers!

Answer 3

Over 512 samples at 44.1 khz might be too incomfortable for real-time feeling. So let's say 80 milliseconds. If you substract this to the local computer buffering constraint (lets say 40 ms) it lets you half the time for network latency.