Portal Setup for Machine Learning course. - MCT-master/portal GitHub Wiki

MCT 4047 Music and Machine Learning – Autumn 2021 setup

Task Overview

The following page gives a detailed description of the setup used in the UiO and NTNU portal rooms for a hybrid conferencing using Zoom and LOLA platforms, for the Machine Learning course taught by Stefano Fasciani.

The scenario for the setup is as follows: The lecturer is physically present in the UiO portal delivering the presentation, 4 to 8 students are also present at the UiO portal. 2 students are in the NTNU Portal, 6 to 10 students are connected in through Zoom.

A solution for the above scenario is implemented by the MCT students, divided into two sections: i) Video setup ii) Audio Setup

Video Setup

The video setup for this course was fairly straight-forward. The MCT Portal Zoom Room was used for video communication. The Logitech Camera was connected directly to Zoom and gave a perspective from the front of the room of all the people.

We did not use the PTZ camera, connected to Zoom through the Black Magic video presenter as a second camera. We tried, but when there would be shared content, the second camera feed would disappear.

To ensure that everybody attending virtually had a good view of the lecturer, a separate Canon camera has been setup. From the Canon camera we took the signal through a SDI-HDMI converter and then connected through an HDMI cable to the presenter’s laptop.

Machine Learning Video Setup Diagram

Problems

Even having a separate camera for the lecturer, the audio was recognized as coming from the main Zoom video, so participants had to manually pin the lecturer’s video. A possible solution would be to search if Zoom has any options of setting a specific camera as the “main” camera.

Audio Setup

In the audio setup of MCT4047, we used the MCT portal as a transit hub to connect NTNU Portal and external Zoom users. Specifically, we used a low-latency, high-quality LOLA audio network to connect NTNU to the UiO portal, and set Zoom audio to connect the UiO Portal to external participants for flexibility and compatibility.

From the perspective of channel routing, in addition to sending the all the audio signal from the UiO Portal to both Zoom and LOLA, we also sending the LOLA signal from the NTNU portal to Zoom and routed Zoom signal from external participants back to NTNU through LOLA, and monitoring LOLA and Zoom at the same time in the UiO portal. In this case, it’s essential to mute local microphone monitoring in both portals and apply appropriate EQ on every channel to avoid feedback.

Since more than half of the students and teachers are physically in the UIO portal, it is important to ensure that the voices of the teachers and each student are clearly presented. Therefore, we adjusted the angle of the AKG C414 microphone in the UiO portal to better cover the students at the far end of the table, and set up a podium microphone for the teacher with higher-level amplifier gain to ensure that the teacher's voice is always presented as the "main voice".

We also prepared a mini-Jack connected to the Midas Mixer and redirected on Zoom for the presenter.

To access the setup, please select the "ML" preset scene in the UiO Portal Midas mixer.

Machine Learning Audio Setup Diagram