Setup for Simultaneous Interpretation of Talks

If your channel would like the c3lingo team of interpreters to translate talks into other languages, please get in touch through Rocketchat #c3lingo or hello@c3lingo.org, and set up your production tech in the manner described on this page.

In order for the interpreters to be able to do their work, two things need to be set up: audio and video from the studio/channel needs to be sent to the interpreters, and audio from the interpreters needs to be transported to the studio to be inserted into the stream.

The studio/channel should provide a stream or video conference where interpreters can hear and see everything that is going out over the stream. Ideally, they can also see and hear what is about to happen. In previous events, being able to see the video mixer multiview and hear the monitor sound (instead of the final mix) was really helpful.

The stream should either be viewable with common desktop clients (VLC, ffplay, …), or should work in a browser. Both low-latency RTMP streams as well as a Jitsi or BBB have been used successfully.

If you're planning to use Jitsi or BBB for production, consider adding the interpreters to that Jitsi/BBB as well. In that case, they would only listen there, but would be muted and have their camera turned off.

If you're using a local setup with an audio mixer, you can also feed the floor audio into mumble, since mumble automatically provides an N-1 setup (the client will not hear themselves).

c3lingo prefers if you take the translated audio from mumble.c3lingo.org. This ensures that all interpreters can have a tested setup, and that they can switch between channels/stages easily. c3lingo will set up mumble channels for your channel/stdio as needed. On the c3lingo mumble server, all translators have appropriate access rights, so that random visitors will not be able to speak in the mumble channels.

The translated audio channels should be set up so that the original audio is audible whenever the interpreters are not speaking. In past events, an automatic ducking setup with an ffmpeg filter complex has worked really well for that. This setup allows interpreters to leave parts of the talk untranslated, for example, when the speaker is reading out a quote in the language that is being translated to, or when a video is being played.

The recording of the translated audio should have no original audio mixed in; for the recording, the mixing will happen in the post-processing stage.

For many talks, it is extremely helpful for the interpreters to review the slides ahead of time, for example to make sure that specific words and terminology are translated correctly. Please consider allowing c3lingo people access to the system where speakers upload their slides.

Alternatively, c3lingo is running their own system where speakers can upload slides and notes.