Subvocal recognition (wikipedia.org/wiki/Subvocal_recognition) - it seems like it's a mostly solved problem in labs, but hasn't been commercialized. I feel like being able to do voice control silently would be a major paradigm shift in how we interact with computers.
Something that hasn't been tried as far as I know is to design a new human language from the ground up with only phonemes that are easily distinguishable when spoken silently.
That is, first we figure the kind of signals that are well separated by ML interpreting EMG output on the throat when subvocalizing a range of sounds, and then we build a conlang using only the precursors to these signals.
Possibly this conlang could then be parsed and translated and the recipient of the message could hear their native tongue instead.
It could start as a computational language at first, only used to send commands and control to software, but then it could be expanded with more nuances and human-centered constructs.
My VR headset has voice recognition. The problem? I talk to myself when gaming and trying to figure something out. So I end up "opening" random programs when I say, "huh, I guess I need to open that next." Detecting that I want something with this technology is the "easy part," detecting that I actually want something right this second and I want actually desire for the device to commit, is the hard part.
I work on voice control with a custom speech engine and models (Talon). I know nothing about subvocal mic hardware or even how the audio sounds, but if anyone has a viable answer to this, especially with a good hardware option, please reach out to me.
Basically, if someone can point me to a solid subvocal speech recognition tech demo and a reliable subvocal mic (note the sibling comment that indicates generic subvocal mics are frustrating and fall out of alignment easily): I think I could train production subvocal speech models, tune support for the audio style in my app, and ship to the public in a matter of weeks.
Consumer throat mics in the field get generally negative reviews. In particular if the transducers shift position even a little, the speaker become unintelligible. That ergonomics problem needs to be solved first.
I think the military throat mics are usually built into flight suits, so they may be less prone to shifting? I know in the lab, they are usually taped into place, which isn't really practical for the real-world.
