Testing Jamulus for virtual rehearsals

[quantum]

I have been looking into methods of doing virtual rehearsals, and Jamulus looks promising.  One of the challenges of common use video conference platforms is latency, they are fine for voice meetings where exact timing is not critical, not so much for music collaboration.  I am working at some ensemble music at the moment, and it would be beneficial to at least do some form of rehearsing while we are still all physically apart. 

With regards to  v3.6.1

What I like about Jamulus:
free
open source
cross platform (Windows, macOS, Linux)
no sign-up needed
ability to setup private server
with private server, data can remain private between people in your ensemble
latency can be brought down to very usable levels
works very well with an audio interface
the UI is clear and uncluttered, accessible for non techy people


What can be improved:
no video as of v3.6.1
buffer and latency tuning options seem a bit coarse to me
could use finer control over audio compression
limited to two audio channels per Jamulus instance


Changing the number of audio channels and the audio quality changes the stream data rate.  So these settings are dependent on your home network and internet connection speed. 

The following video is a test comparing the audio quality settings in Jamulus.  There is a very noticeable difference between Low and Normal.  I think more granular control over the audio compression would help better tune the system to an end users hardware.  The jitter buffers were set manually for best sound, however, you can still hear dropouts in the audio.  Despite the dropouts, I still found the experience usable from a practical perspective.  All devices used wired network and wired audio connectivity. 

[quantum]

How is the latency?  For the following test I wanted to isolate some of the variability associated with playing across the internet.  It was carried out using the local network and a private Jamulus server. 

The test was carried out using Jamulus 3.6.1.

Buffer Delay was set at 64 samples, the lowest setting displayed in Jamulus.  Although, it is possible to go even lower if you modify the buffer from inside your audio driver rather than the Jamulus settings window.  These settings resulted in an overall delay of about 48 ms. 

Piano solo part was recorded first, as it is more active, and takes the lead in the music excerpt.  The objective was to play the organ part to a playback recording of the piano part through Jamulus.  Naturally, this is all much easier to accomplish in a DAW, but that is not the subject of this test.  The 48 ms delay was very manageable to me, and I was able to quickly adapt.  However, there were some things that were observed in this test that are worth noting. 

It is difficult to judge musical balance through Jamulus.  The faders in Jamulus determine the balance and mix.  These faders appear to be independent to each client, meaning, each person can set their own faders.  It is fine for achieving a personal mix, but gives little insight towards the ensemble balance.  In certain types of ensemble music, musicians depend on the aural feedback of the ensemble sound to adjust their playing - this is difficult with the way Jamulus is setup. 

In my case, I also found that the monitoring sound through Jamulus, for monitoring ones own instrument as it is being played, was very different from the overall sound levels captured on the recording computer.  For the test, the monitoring sound was much softer compared to what was recorded from Jamulus after the sound had passed through it.  All the faders were set at 100%, yet there was still a difference in sound level. 

Regarding reverb, the one provided with Jamulus is simple.  No ability to control the dry/wet mix.  If you need reverb in your sound, it is probably better to use an external reverb.  In the test case, the organ samples used were expected to be used alongside reverb, however, I was not completely satisfied with how they worked using the Jamulus reverb.  The reason behind choosing a dry sampled instrument was to avoid a mismatch in acoustic between my home piano which is in a room with relatively dry acoustic. 

As for making a recording from Jamulus for an ensemble, there are still too many dropouts in this test.  I may find it a better proposition to record each person individually then assemble later in a DAW.  In such case, Jamulus would be used for syncing the music, not as a recording source.  It may just be a matter of further tuning the buffers or network.