How to Rehearse a Choir Online — and Actually Hear the Ensemble

Yes, a choir can rehearse online. But conventional videoconferencing is poorly suited to ensemble singing, and any director who has tried it already knows why.

Internet delay is not the same for every person on the call. Your sopranos in one city and your basses in another arrive at each other at different times. So Zoom, Teams and Meet do the only sensible thing available to them. They mute everyone except one singer or one accompanist and make people take turns. What is left is not a rehearsal. It is somewhere between a conversation and a broadcast with a chat window.

There are several different approaches to remote choral work. Some systems help singers learn their parts alone. Low-latency music systems try to squeeze Internet delay down far enough that musicians can react to one another directly. Lyrekos takes a third approach: every singer performs against a common synchronized reference, which allows their voices to be combined in sync even when they are thousands of miles apart.

The right approach depends entirely on what you mean by “rehearse online.”

Why can’t a choir just rehearse on Zoom?

Because ensemble singing is a timing problem, and videoconferencing was built to solve a conversation problem.

Here is the chain. A singer hears the accompaniment. She sings a note. That sound is captured, encoded, sent across the Internet, decoded and played to a second singer. He reacts to what he hears and sings. His note makes the same trip back. Every leg adds delay. What matters most is the roundtrip time.

For conversation that is fine. A half-second pause before someone answers a question is ordinary human behavior. For ensemble music it is fatal. Two singers trying to be in sync while each hears the other a quarter-second late will pull each other apart within a bar, and the tempo will sag while they do it. Every choir that has tried this has discovered the same thing: the ensemble does not merely sound bad; it is a cacophony that actively decelerates.

Videoconferencing platforms are not doing anything wrong. They are optimized for speech intelligibility over unpredictable networks. This is why they compress aggressively, suppress what they classify as background noise, cancel long-held notes because they think it is an echo, and hand the floor to one talker at a time. Those are the correct choices for a staff meeting. They are the wrong ones for a song or even a chant.

How much delay can singers tolerate?

Roughly 20 to 30 milliseconds. The answer is that it depends on tempo, texture, and what the musicians are being asked to do, but that range is the working figure most people in this field use. It sometimes needs to be a little tighter when musical instruments are involved.

Use the room you already rehearse in as a ruler. Sound travels about one foot per millisecond. The singer standing twenty feet away across the risers is already reaching your ears twenty milliseconds late, and you have never once thought about it. That is your listening circle, and choirs live inside it comfortably. We walk through where that threshold comes from in more detail elsewhere.

Now push the number. At sixty or eighty milliseconds the ensemble starts to drag. Past a hundred, singers can no longer hold together at all, and no amount of discipline fixes it, because each person is faithfully following what they hear.

What the delays actually are

Here are numbers we measured during our Australia to Los Angeles test. Our cloud server was in Oregon. These are roundtrip figures. They include the network and the browser but not the jitter buffers that absorb variation in delay, and those buffers add significantly.

Compare those to the 20 to 30 millisecond budget. They are not close, and they are lower bounds. Note that even the short hop, two neighborhoods in the same metropolitan area, misses by a factor of ten.

There are ways to attack this. If everyone were in Los Angeles, you would abandon the browser and use a specially configured downloadable app. You would move the server into the city or eliminate it. You would ban Wi-Fi and require Ethernet. We wanted to work worldwide and we wanted it to work without IT support, so we went a different way.

The full breakdown — why the browser costs what it does, what the jitter buffers buy, and the four ways anyone can attack the problem — is in why Zoom fails for music.

Why does online choir singing get out of sync?

Because the delays are unequal and they change.

If every singer were exactly 40 milliseconds from every other singer, a choir might be able to adapt with enough extra practice. The human brain is amazing. But one member is on fiber, another is on a cable modem two hops from a congested exchange, and a third is on hotel Wi-Fi. Their delays differ from each other, and each person’s delay wanders during the rehearsal as networks reroute and buffers refill.

So the ensemble does not just start late. It drifts. And because everyone is reacting to a slightly different picture of “now,” there is no shared beat for anyone to return to.

This is the observation that the rest of this article is built on. Fixing remote choral rehearsal is not primarily a matter of making the Internet faster. It is a matter of deciding what must happen in real time and what does not.

Three different things get called “online choir rehearsal”

A great deal of confusion in this subject comes from the fact that one phrase covers three unrelated technologies. It is worth separating them before you choose anything.

1. Individual part practice

The soprano practices against a soprano-dominant track. The tenor practices against a tenor track. Products in this category are mature and useful, and for many ensembles they are the highest-value purchase, because most singers’ real problem is not the rehearsal, it is the six days between rehearsals.

But nobody is rehearsing together. One person is singing along with a recording. The director is not in the room, cannot hear the result, and cannot change anything.

2. Low-latency networked performance

Systems such as JackTrip, Jamulus and FarPlay attack the delay directly. They strip out everything that adds milliseconds: minimal buffering, lightweight or no compression, wired connections, dedicated audio interfaces. Under favorable conditions the results are excellent, and for two or three musicians who are geographically close, technically capable, and properly equipped, this is a real solution to real-time playing, especially for free-form jamming.

The constraints are the constraints of physics and of setup. Distance still costs milliseconds that no software can recover. And the configuration burden falls on every participant, not just the organizer. That is manageable for a jazz trio of engineers. It is a different proposition for forty volunteer choristers with an average of one technically confident person among ten.

3. Synchronized distributed rehearsal

Everyone performs against a shared timing reference rather than against each other. Instead of demanding that the Internet deliver every singer’s voice to every other singer almost instantaneously, the system keeps every performance aligned to that common reference and combines them in sync afterward.

This is the approach Lyrekos uses. It is the reason singers in California and Australia can end up in the same ensemble, which the low-latency approach cannot do over the Internet at that distance no matter how good the software is.

It carries a real constraint. There must be a reference.

What actually happens in a synchronized rehearsal

This is the part most explanations skip, so here it is concretely.

What does each singer hear?

Headphones — open-backed are best, because the singer needs to hear their own voice — and the shared reference.

The reference might be an accompaniment track, a rehearsal piano, a previous pass by the full ensemble, or a strong section. Lyrekos records each input on a separate track, so there is great flexibility in building toward the final performance.

For the California to Australia test of Amazing Grace, we did the first take with a prerecorded piano backing track. Everyone heard the piano and sang their part to it. Note that they didn’t immediately hear each other. As soon as the take was over, we played the combined track back to everyone. Near instant feedback. It was OK, but the singers missed hearing each other.

We then did a second take. This time we used the first take as the backing track. Thus each person heard everyone, and they were able to sing together. Note that it was a little bit of a trick, since what everyone was hearing was the previous take, but we were delighted to learn that it was still a compelling experience and a good emulation of what happens in a real choir rehearsal or performance.

The director who sang in this distributed session described it this way:

It emulated an actual choir experience where you listen to certain voices doing certain things and you match them. The alignment with all our consonants came together in that second take because everyone heard what I was doing.

Alex Siegers, choir director

That is the mechanism working as intended. Singers were not matching a grid. They were matching each other.

We could then, of course, have used that second take as a backing track for a third take, and so on. If you listen to the recording, you will note that you don’t hear the piano. The result is a cappella. This was possible because, as we mentioned earlier, each track is recorded separately, and so we were able to suppress the original piano track by choice. Notice the flexibility this gives one. The backing track could have been a famous singer or choir that one wanted to make one’s own.

What does the director hear?

The assembled ensemble, in sync, as a choir. This could be during the take if the director is not themselves singing, or immediately afterward if they are.

That is the whole point, and it distinguishes Lyrekos from a virtual-choir video project. In a virtual choir the director hears the result days later, after a sound engineer has assembled it. Here the director hears the ensemble during the rehearsal, which means the director can make musical decisions during the rehearsal. Balance, blend, diction, tuning of a specific chord. The ordinary work.

Can we work on measures 23 through 40?

Yes. This is normal rehearsal behavior.

The director sets the passage, the ensemble sings it, and the director runs it again. The distinction between a rehearsal tool and a recording tool is exactly this: whether repeating a difficult passage eleven times is a chore or the native activity. It is not quite as fast as it would be if everyone was in the same room, but it is a lot faster than flying from Sydney to Los Angeles.

Can I rehearse only the altos?

Yes, and this turns out to be one of the strongest reasons to use a remote tool at all.

A sectional is the rehearsal that is hardest to schedule in physical life because it asks eight people to travel for thirty minutes of work. Remove the travel and the sectional becomes cheap. A director who could never justify calling the altos in on a Tuesday can now do it in the time it takes to send a link and grab 40 minutes together online.

Can singers hear the other voice parts?

Yes, and they should. Part-independence is learned by hearing the parts you are not singing. A system that isolates each singer with a click track teaches choristers to sing accurately and to listen badly.

Does this help with difficult repertoire?

This turned out to be the use case singers raised first, unprompted, and it is a different argument from the scheduling one.

Practicing a hard piece alone teaches you your line in isolation. The problem arrives in the room, when your line must be sung against the three lines it clashes with. One singer in our sessions put it plainly:

For difficult ensemble music, one per part, eight voices, it’s quite intricate. You rehearse it alone, you listen to recordings, and then you get in the room and it’s “oh God, I have to sing right into this clash, I have to sing a tritone up.” I think this would be super useful.

Cody Christopher, singer

The value is that you can sing your part alone, then immediately sing it against the other parts, then again, on a Tuesday, without booking anyone’s evening. Part-practice tracks cannot do the second half of that. A room can, once a week.

Can I hear an individual singer?

Yes. Each singer is recorded on a separate track and can be reviewed immediately, or at the director’s leisure any time after.

Can the director actually direct?

Yes, in a couple of ways. Like in a rehearsal room, it is easy to have a conversation. The director can say what she wants and people hear what she says, just like any videoconferencing system.

But more than that, the director can direct. We have a mode where all the singers can hear and see the director as she directs, including hand motions and lip movements. Of course, they are hearing and seeing her a few seconds delayed, but that doesn’t matter to them, and the system syncs it all back together ready for instant review.

The reason this works is that direction travels one way. A gesture only has to reach the singers. It does not have to make a round trip and come back inside a beat, which is what would put it up against the timing budget. What the director cannot do is respond to something a singer does inside the same take, because that would be a round trip. She hears it immediately afterward instead.

From one of our sessions:

As a director, that was good to be able to say, hey this is what’s happening, and then people being able to follow it.

Alex Siegers, choir director

Can we listen back afterward?

Yes. Because every performance is already aligned to a common reference, a synchronized recording is a byproduct of rehearsing, stem by stem, not a separate production step.

Example: a Tuesday evening alto sectional

Abstractions are less useful than a scenario, so here is one.

A community choir rehearses Sunday afternoons. Two passages in the second movement are not working, and the problem is in the altos. In a normal week the director has two options: spend twenty minutes of full-choir time on eight people while sixty others sit, or just let it go.

Instead, the director sends the altos a link on Tuesday morning. At eight in the evening, eight people open it on laptops and phones, put on headphones, and work the two passages for half an hour against the ensemble reference. Nobody drives anywhere. Nobody arranges childcare. No one deals with Tuesday rush-hour traffic. The director hears the section in sync, fixes the entrance in measure 31, runs it four more times, and everyone is off by 8:35.

Sunday’s rehearsal now starts with those passages already solved.

The value here is not that the technology is impressive. It is that a rehearsal which was previously impossible to schedule became a thirty-minute decision.

How far apart can singers be?

We have tested about eight thousand miles. We have also tested over Starlink.

In our demonstration, four singers performed together from Los Angeles, Canberra and Melbourne over ordinary consumer Internet connections, in a browser, over Wi-Fi, with no special hardware and nothing installed. The recording is on our blog: an arrangement of Amazing Grace.

That distance is not a marketing number. It is the experiment that demonstrates the architecture. California to Australia is far past the point where the low-latency approach can work, so a working four-part ensemble across that gap is evidence that the system is not quietly depending on low latency after all.

The singers’ own reaction is more useful than our description of it. One said it felt together and sounded genuinely good. Another said that stepping back and remembering they were spread across the world was the striking part, because it had not felt like it while they were singing.

If it works at eight thousand miles, the choir member who moved to Denver is not a hard case.

What do singers actually need?

A browser, headphones, and a reasonably current phone or laptop. If one’s Internet is good enough for Zoom, it is almost guaranteed to be good enough for Lyrekos.

That is the whole list. And the director doesn’t become the IT help desk.

Wi-Fi is acceptable. Ethernet is not required. Nothing to install. If a singer can join a video call, they can join a rehearsal.

What this approach does not do

Worth being plain about the edges, because a tool you misunderstand will disappoint you.

• A singer cannot react to another singer’s unscripted change inside a take. Reaction is a round trip, and a round trip across distance does not fit inside the timing budget. Direction works because it only goes one way. If I can hear you, you cannot hear me in time, and vice versa.
• There must be a reference. For a choir working on repertoire, one usually already exists or can be made in a single take.
• It does not replace the room. Singing in your closet is not the same as standing next to forty people in a live acoustic, and we are not claiming it is. If you can get together in person, do it.
• It does not fix unprepared singers, but it makes it much less hassle to rehearse them.

What it changes is how often a choir can work together. For most ensembles, travel is the constraint that binds.

Where to start

If your problem is that singers arrive not knowing their notes, buy part-practice tracks. That is a different problem, and it has good solutions.

If your problem is that you can only get everyone in a room once a week and it is not enough, that is the problem we are purpose-built for.

Lyrekos runs in a browser, and we are working directly with choirs and directors while the product matures. If you have an ensemble and a rehearsal you cannot schedule, tell us about it.