Playing together

How to play music with someone online, in real time

Two musicians can genuinely play together over the internet, and the thing that decides whether it works is distance, not software. Light in fibre costs about 1 ms of round trip per 100 km, and no product repeals that. Inside a city or a neighbouring one it feels like playing together. Across an ocean it does not, and any service that tells you otherwise is selling something.

By MELD · 12 September 2026 · 8 min read

What is the actual limit?

Start with the part nobody can optimise. ITU-T G.114 puts light in optical fibre at 5 microseconds per kilometre. So 100 km one way is half a millisecond, and the round trip — you to them and back — is about 1 ms per 100 km.

That is the floor, and real connections do not achieve it. Fibre does not run in straight lines between two houses, so multiply by roughly 1.3 to 1.5 for the route a packet actually takes. Then add the switches along the way, the jitter buffer that smooths out uneven arrival, the codec at each end, and your own sound card in and out.

A useful rule of thumb from those numbers: within about 500 km, physics is costing you well under 10 ms of round trip and everything else you feel is fixable. Across 5,000 km it is costing you 50 ms or more before any equipment is involved, and nothing you buy will change that.

How much delay can musicians actually tolerate?

This is where most articles quote a single confident number. The research does not support one, and the disagreement is more useful than the number.

• 20–30 ms one way. The classic figure, from Nathan Schuett's 2002 CCRMA thesis, which coined the term Ensemble Performance Threshold — the one-way delay below which players can still synchronise. Almost every figure you see quoted traces back here.
• **11.5 ms one way is the optimum.** Chafe, Gurevich, Leslie and Tyan (2004) ran clapping duets at controlled delays and found an accuracy sweet spot at 11.5 ms. At 14 ms and beyond, 85% of performances slowed down.
• At least 40 ms one way. A 2025 study by Tsioutas, Xylomenos and Doumanis argues the older figure is too strict, finding that musicians on real instruments synchronise at delays up to 40 ms, "compared to the 25–30 ms reported in rhythmic hand-clapping experiments".

The Chafe finding has a detail worth keeping, because it is counter-intuitive and it is properly measured: too little delay makes ensembles rush. Below 11.5 ms, 74% of the performances accelerated. Musicians are used to a small amount of acoustic delay — a drummer three metres away is already about 9 ms of it — and removing it entirely is its own kind of wrong.

Our own experience matches the more forgiving end. Playing over a link measured at 80–90 ms round trip — roughly 40–45 ms each way — it still sounded clear and immediate to both players. That is one pair of musicians, not a study, and we report it as such.

So how far away can your bandmate be?

Do the arithmetic with the numbers above rather than trusting a marketing claim.

And treat the long distances as an opportunity rather than a failure. Call-and-response writing over a slow link — one plays a phrase, the other answers — has produced real songs. What does not work is pretending it is the same thing as being in a room.

What makes it worse than physics says?

Usually the room you are sitting in, not the internet between you.

• Bluetooth headphones. The single biggest avoidable cost, and it is not close. Wireless audio adds enough delay on its own to push you past every threshold in this article. Use wired headphones. This is the first thing to change and it is free.
• Wi-Fi. Not so much the average delay as the variability, which forces a bigger jitter buffer, which is pure added latency. A cable to the router is worth more than any setting.
• Echo cancellation and automatic gain. Designed for meetings, ruinous for music: they duck and pump a performance and add processing delay. A tool built for playing turns them off.
• Video. Sharing a picture competes for the same connection as the audio. Audio first, always; a stuttering video is an annoyance, stuttering audio ends the session.

What actually helps

1. Wired headphones, wired network. Both ends. This alone rescues most sessions.
2. Play with people near you when you want to play. Save the distant collaborators for writing and for sending parts back and forth.
3. Turn off anything that processes the signal — echo cancellation, noise suppression, automatic gain.
4. Agree on a count-in. Two people at slightly different delays need a shared reference more than two people in a room do.
5. Look at the number. A tool that shows you the real round trip lets you know whether to change something or accept it. A tool that hides it is hiding bad news.

How MELD Jam handles it

MELD Jam puts you in a room with another musician in a browser tab. No account, no install, no plugin: pick what you play and what you are looking for, or send someone a link and the two of you get a private room.

It measures the round trip in the first second and shows you the number rather than hiding it. If the connection is worse than 100 ms round trip you get told, with two buttons — keep looking for someone nearer, or play anyway. Nobody gets dropped into a session that cannot keep time and left to conclude that the platform sounds bad.

It records while you play, your own microphone only, and the take goes to the Studio timeline, to a download, or straight to a post. Recording both players is a switch each of you turns on before being matched, and it only ever pairs you with people who turned it on too.

One honest caveat: MELD is early and the public lobby is often empty. The invite link is the reliable way in — it always finds the person you send it to.

What you do with the take afterwards is the other half: building it up take over take in a browser, then mastering it to a loudness that survives streaming.