In this topic, I’d like to welcome any discussion about how best to protect privacy and security when sharing pictures/videos online and live-streaming video to an audience, as well as necessary compromises that sacrifice privacy and/or security to some extent. Relevant product recommendations for things like cameras, tripods, and lights are also welcome.
I made a separate topic specifically about metadata here:
Librem 5 (and standalone cameras) photo metadata
Maybe YouTube is evil and I should never use it, but if we save morality for another topic and consider for the time being only security and data privacy, I’ll admit that over the year and a half as I tried to move more to Purism hardware and PureOS where possible, I never totally stopped making and uploading YouTube videos on my channel with about 1000 subscribers. They tell me that I don’t serve the algorithm well enough to enable monetization, so I am quite simply freeloading off of their system for its communication functions and not using it for any financial gain.
But when I make a video tutorial about how to use my software, or giving a demonstration of results that might be created with it – or in some cases just ranting and reflecting on life – a lot of times recently I find it more enjoyable to fill in the voice with an AI generated voice instead of my own. Even if the data entry of what I want the AI to say is more easily achieved by using a mic and then using some AI-based converter to generate audio in a different voice using the same words and intonation as what I originally spoke in the mic, it’s getting to the point where one way or the other making AI generated audio is well within the realm of capability in our society.
Now, perhaps I have misled myself to some degree on that topic, because I was using a nonfree JavaScript web login to do the generation. And realistically that’s stupid and I came to realize I should stop supporting that. It would be much better for society if AI generation software was free software that any of us could load on our own computers. For one of my videos just ranting and being silly, the site that I was using as a paid service said that even if I pay them, they prevent generating audio in the likeness of a certain political figure because someone decided that shouldn’t be allowed. For me that was a reminder of how solving problems with stupid online portals instead of our own machines is fated to end up with bizarre nonsense limitations that have nothing to do with technical limitations.
But these points of ethical concerns aside, they say the latest human technology can generate anything in anyone’s voice if we have 3 seconds of audio in that voice. But my YouTube channel has dozens of hours of my voice. So, if I had the time, I would probably redownload all these videos and slap some different AI generated voices on them, and re-upload.
Already when I was doing TTS for new videos, some videos I would use a voice like Emma Watson, sometimes I would Agent Smith from the Matrix. When the purpose of the video is to make some funny comment about niche technology or perhaps teach how to use software that I have written, it doesn’t matter if sometimes I’m Agent Smith and other times I’m Emma Watson and other times I’m Furion Stormrage. But by doing that, I can speak to the internet on a site like that without leaking personal data. So if I was going to be intentional about it, I really think I would nuke the entire YouTube channel and then only re-upload things with AI voices over top that did not leak audio/visual personal data about my real life. It’s basically too late for me, but if it weren’t, I think that’s what I would do.
Edit: Here are some examples. These were sufficiently easy to make that I don’t really think I should use my actual voice anymore:
(generic female voice):
(synthetic discussion between generic male voices):
(Agent Smith from the Matrix voice):
(Emma Watson):
(Mosaic of many voices):
But in all these cases, I can create and upload this content without needing to expose my own human voice to the outside.
You can use a Peertube (https://joinpeertube.org) instance, or host one yourself, and it offers video hosting and live streaming. Peertube is a Youtube and Vimeo alternative and is open-source.
One Peertube instance is TILvids (https://tilvids.com/) and they focus on edutainment, that is, videos that are both educational and entertaining. They promise not to collect any additional data outside of what the underlying software (PeerTube) may collect.
What user information do you collect?
Respecting data-privacy is one of the core foundations of TILvids. We utilize the open-source PeerTube software ecosystem to run the site, and collect nothing more than what PeerTube collects out of the box.
If you just want a live streaming, self-hosted solution, Owncast (https://owncast.online/) only offers live streaming. Owncast is a Twitch alternative and is open-source. As of 2024-04-20, there is no option to record VoDs from a livestream currently.
For audio, I would suggest using an audio interface like a Focusrite Scarlet 2i2 or Motu M2. I am not endorsing either of these and am just providing examples. Typically, microphones interface with them with XLR so you might need to purchase a microphone on top of the interface if you don’t already have a compatible microphone. The main advantage of using an audio interface is that you have more control over the gain instead of doing it through software.
If that is outside of your budget, you can consider a more affordable USB microphone like the Blue Yeti to start with.
I highly suggest you obtain a pop filter with whatever microphone you use so that words with hard consonants, such as “pop” (hard “p”) or “train” (hard “t”), are softened for listeners.