They refer to four approaches to multilanguage gatherings and map out a plan for running session#4 in commons.hour (led by femProcomuns) as a dual-language Spanish/English session. @camille might be interested, as well as @wouter@benjaminroyer and Cade (forgotten his Forum handle, oops!)?
Contributors in this thread may be interested in the discussion thread on commons.hour #9 Working in plural languages online Session#9 discussion - #3 by mikemh
It’s mostly a matter of meeting protocols. There isn’t any specific tech fix for simultaneous interpretation - like an interpreters’ control panel for example, as used in multilanguage conferences witth special audio equipment. Also we’re thinking about machine translation in various forms, but nothing specific on the roadmap I think @wouter ?
From many members in our community we know that multi-language is very much desired, and we are aware of the current difficulties. Still, as @otisyves, @fredsultan and Matthieu from Remix the Commons have explained so well in the last #commons.hour it depends very much on the social protocol how to go about this. But tech can be more helpful. Add also feedback from users in loud environments, like a general assembly or meetings in public spaces: automatic subtitling would be very practical.
Based on these user requests we do have “improved multi-language features” on our roadmap. It is however to be defined in more detail about what it is that we want, and then we should seek funding, possibly crowd-funding, for it.
I could imagine this automatic subtitling to be very useful and possibly also the most feasible. And possibly added to that automatic translation of these subtitles.
We’re probably not the first to attempt this, although I haven’t found public repositories under free licenses that we can reuse so far yet.
One speech-2-text tool that @mikemh pointed me to is nerd-dictation. Probably @baybars knows much much more of this domain. Would be good to explore this together.
Hey @wouter!
Just finished some quick tests with VOSK (VOSK Offline Speech Recognition API) - the Open Source library under nerd-dictation. I used few audio files from Remix collection. The results are quite impressive (in terms of accuracy - speakers from France and Morocco - I didn’t test with a French-Canadian speaking person…). The library seems to integrate well with other systems.
It doesn’t provide translation but a « live » transcription - a text (pretty rough one) that can be translated afterwards. That’s a start.
The prospect of improved multi-language features, including automatic subtitling and translation, sounds promising. Considering the importance of user feedback and potential benefits in loud environments, I’d be interested in exploring this further. Regarding speech-to-text tools, nerd-dictation is a good start, and I’m open to collaborating with @baybars and others to delve deeper into this domain.