EasyEffects can change mic input live. Privacy option for people who need to use their voice in any public/insecure online setting & don't want to be easily identified.
(github.com)
from communism@lemmy.ml to privacy@lemmy.ml on 02 Aug 2024 00:58
https://lemmy.ml/post/18672078
from communism@lemmy.ml to privacy@lemmy.ml on 02 Aug 2024 00:58
https://lemmy.ml/post/18672078
There are other FOSS real-time voice changers for Linux, but the others I found either seemed to have fewer features, be less polished, or be abandoned.
I’m not really a voice expert or anything so I’m not sure what aspects of voice a, like, forensic voice analyst or something would look at. I’ve just changed the pitch and I sound different enough that I wouldn’t recognise the voice, which is good enough for me. Open to suggestions as to what effects would give the most privacy in terms of making it harder to identify your voice (while still being intelligible)
Also, for people’s reference, if you want mic input to be changed for all apps, go to three dots > Preferences > General > Audio > Process All Input Streams and enable.
#privacy
threaded - newest
STT | TTS > /dev/audio ?
Hey that’s what I did with Mimic2 (the Mycroft project’s TTS) and DeepSpeach (Mozilla’s STT) for a Lil bit. Hard to handle streaming data right, but it was doableish.
I tried it last year, but didn’t find any presets that were effective and practical for voice disguising, and didn’t have time to learn how to wrangle the low-level effects to do the same.
(To be clear, pitch shifting is trivial to reverse, so not what I would consider effective or practical for this purpose.)
It was easy to set up, though, and appeared to be better maintained than some of the alternatives. I might give it another try eventually. Perhaps some commercial audio plugins would help.
Yeah that’s fair. And my pitch-shifting is not for a particularly well-resourced/dedicated threat model, literally just “would I recognise this as my voice if I heard it”—obviously insufficient if your adversary is going to put effort into identifying your voice.
I’m a bit surprised that no one has made a good preset for voice disguising. I might just keep layering effects for further obfuscation.
Voice disguising is one thing AI can actually be useful for. Replacing the voice altogether with a modeled one will be a better disguise than applying well known audio effects. Then you just have to make sure your vocabulary and phrasing don’t give you away.
I personally don’t need sophistication, I just want to not sound like a woman in a game voice chat. I wonder if this is what I need.
I hope the day will come soon when there’s no longer a reason to avoid sounding like a woman in game chat. Hearing only male voices gets old fast.
Interesting project for an issue that I feel like never gets talked about. No idea how easy or difficult it would be, but if the dev could build in VST support to the program that would give this project nearly unlimited potential.
In the meantime, its possible to use qjackctl to create a connection from input to the VST before it goes to the easyeffects sink. Its a bit kludgy but it should work well enough.
A must-have in my setup, love it
One of the few practical things AI might be good at:
github.com/CorentinJ/Real-Time-Voice-Cloning
I’ve found EasyEffects also useful for boosting volume on sites with quiet video. Set up a hotkey to enable a compressor with gain when that happens.
Thank you a lot! was looking for smth like that for privacy reasons.