There are 'global phases', first for description/imagination and the second for execution/comparison.
Everyone has a sheet with a table for noting estimated time and one sheet is at the table for real time.
One by one from the starting player people describe their ideas for others who estimate and note down their bets (hidden betting although engaging in a misleading banter is a good fit here ;) ).
Put your sheets down still hidden.
Then music starts and it runs in form of "movements" of number equal to the number of players. Time-tracking is done by the last player for the first movement, and by the first player for all the next movements. The real times are written on a public sheet.
Pick up your sheets, compare seconds, announce a winner.

Still lacking is how to describe a description. What actually the game expects and demands from players at this stage is for me now a key matter.

Thinking about it from another level of distance now, what I'm mainly hoping the game does is this:
description
imagination
execution
comparison

I want it to be simple enough that it moves very simply from one to the next. I could imagine stealing some of the mechanics of betting, for example on horse races. (I would need to research these.) Or it could simply be that each person fills in their own chart and any discussion would happen after the actions have been executed. If it were done after a break (allowing for preparation) and each person was responsible for his or her own numbers, there would be minimal down time. Based on whatever happens in the first round, the players could make additional constraints or rules for a second round.

Thanks for thinking it through with me!

Would it be helpful to list some possible (non-musical) actions as prompts?

Helpful although dangerously suggestive. Maybe more elaboration for "how to describe your sound" could do. In our test most common form was similar to a 'long title'. Your "rolling or dropping" example is informative just at this level - as you would describe the action more than sound itself.

It would also be good that the performer not know the guesses in advance.

Huh, we were taking care of that (guess it seemed obvious/natural gaming-wise), but at the same time it might be what extends the downtime.

Another option would be that each player has a pin or a tag or a marker of a different color…

For game it might be better, but as for the contents and connotations this is some kind of a revolt ;). Numerology, reducing sounds to digits, this is strong in the game now, but surely it might be also about space, why not…

one of several different activities…

As for the context, I remember thinking that this is one of 'after-the-break' games. This is quite a common type, where players need to invent or plan something out, so it's best to hint them their task before the break, so they are prepared at the start of the game itself. We didn't do it like that, so there was some waiting with just that.

I think having a Q&A section before the sound might make the guesses less random or luck based.

This might be interesting. Do you have a procedure in mind? There are risks of slowing the flow down and it's hard to quickly introduce some dynamics here, like ways to get informational advantage over other players, trade-offs with the risk of helping others more than yourself etc.

I should clarify (probably within the instructions as well) that the sounds don't have to be music-based, though they could be. It could be any action that makes noise. The performer can have more or less control over the sound, depending on what they choose. I'm not sure how that plays out in terms of strategy. It doesn't have to be skill based, at least for the performer. They could drop or roll something on the floor or a table, for example. I don't think performance skill is a great advantage, since no one is guessing the duration of their own performance. Would it be helpful to list some possible (non-musical) actions as prompts?

About the dead time, I like the idea of each person keeping their own score card. It would also be good that the performer not know the guesses in advance. Another option would be that each player has a pin or a tag or a marker of a different color, and puts their guess onto a timeline score sheet. Then the actual time could be marked on there and you could visually see who came closest. I think I prefer this option, since it gets away from extra calculation time, but I'd be interested in your thoughts. In both cases you would still need a timekeeper.

I was trying to think about incentive structures. For me, this is more of a mini-game or activity, probably not something to be played for an extended period of time on its own, so I'm reluctant to make it more complicated to learn to play. It could stand separately as one of several different activities. Or it could possibly be one component of a larger game.

I think having a Q&A section before the sound might make the guesses less random or luck based. The people who are guessing could ask the performer one question each before committing to a guess. I'm not sure if that would increase interaction or create new timing issues.

The game currently lets different participants play experimental music by making it easy for those that don't like it. You can listen intently, but if you don't want to (or can't) you may just focus on imagined seconds [;)] and the activity gets less awkward for you this way… By distracting the listener it also takes the stress of the player.

And I feel like book-keeping serves as an "attention hook" and it influences how people play and how people listen. Making a series of ultra-short recitals would be socially much different. But indeed I think with some tweaks this activity could be accessible to many more people.

The dynamics of "Imagined seconds" are really interesting. All participants focus for some time just at one player who performs a contained part. That feature is unique, refreshing and also game-wise weird as there is practically no interaction at the core.

It might seem counter-intuitive then to suggest:
1. Moving away from one table filled in by Timekeeper. When everyone has own chart, Downtime should be much lower even with many players, as tables are filled simultaneously. Right now it's considerable, especially when you headline your game as "for any number of players"… ;)
I'm pretty sure this will still not hinder the Interaction. Here it mostly takes the form of comparing the bets, which is indeed done during the betting, but can also be expected after the round and is then not forced but naturally encouraged.

2. Informed choice would be another issue I think of right away. Should the "bets" be as seemingly random as they are now? A natural mechanic would be to provide incentives for the player to describe the sound in sufficient detail (such a mechanic is exemplified in Dixit, when you lose points when no-one guesses, although it's also not good for you if everyone knows the correct answer).

What else?