Since my last post, I’ve been busy preparing for our live singing test with volunteers from Ragazzi’s YME Chorus. We distributed 7 test kits consisting of Raspberry Pi 4B’s with HifiBerry DAC+ ADC, a microphone, all necessary cables, etc. The boxes are designed using a custom image to run headless (no monitor, keyboard, or mouse), and are managed remotely using a new cloud service that I’ve been building out (more about this in the future). The idea being that you shouldn’t have to do much more than plug them in and sing. This makes the low-latency audio technology far more accessible to a broader audience by (a) making it a lot easier to get started, and (b) minimizing hardware/software latency so that it works better across more Internet connections.
We completed our first singing test this past Wednesday. Jamulus seemed to work great for everyone, but JackTrip (48k, 128 and stereo) only worked for a few people. Others heard distracting artifacts or in one case, nothing at all. I suspect their upstream bandwidth couldn’t handle the demands, but rather than trying to troubleshoot on this call, we just switched back to Jamulus. I may try JackTrip using mono with larger buffers next time.
We had 6 boxes connected on the first test: 4 boys singing with an accompanist on piano, and myself listening in. Server load was about 1/4 of a single CPU core, which suggests that a high-end EC2 VM could handle about 192 participants.
I recorded the call using Audacity by connecting the output from my RPI to a Focusrite 2i2. Below are two samples (with added reverb) for your listening pleasure. Unfortunately, I had the input level set too high, so you may hear some clipping on the louder parts. Also note that this is the first time these boys have been able to sing together in several months. Especially considering all of this, I’m thrilled with the results!
We plan to continue testing weekly, while also growing the group of participants throughout the Summer.