I am trying to set up voice control software (talon with a wav2letter backend) to work well with shell completion. I would like to be able to capture what the results of a completion *would* be if it ran, but without the normal showing of results. The goal is to be able to feed the completions into the voice command software so that the next spoken word can be chosen from the available completion options. The nature of the talon/wav2letter combination seems to require that you have a wordlist first and the audio is used to guess which word was most likely matched. So then the challenge is how to expose the current completion options from a zsh prompt. The only prior art I could find was this hack on github: https://github.com/Valodim/zsh-capture-completion But that's not particularly useful to me. I could imagine two high-level ways of doing it: - A "push" strategy where after every keystroke the completion code is updated, a la https://github.com/marlonrichert/zsh-autocomplete. Only in this case the results would be written to a file named after the WINDOWID (this is all under X11). After that, watching for file updates and tracking the focused window and notifying talon accordingly is easy. - A "pull" strategy where some callable function or widget can return the results of a completion (without altering the prompt of course). That function/widget could be triggered by trapping a signal or by reading a packet from a socket or something. After that, I would be able to hook the rest of the system together I believe. I'm pretty new to trying to hack zsh. I spent a bunch of time trying to read docs and make things work but I still don't really have any idea of what I want is possible (or how much coding it would take to become possible). Does anybody have any ideas or pointers for how to do this sort of thing?
I am trying to set up voice control software (talon with a wav2letter backend) to work well with shell completion. I would like to be able to capture what the results of a completion *would* be if it ran, but without the normal showing of results. The goal is to be able to feed the completions into the voice command software so that the next spoken word can be chosen from the available completion options. The nature of the talon/wav2letter combination seems to require that you have a wordlist first and the audio is used to guess which word was most likely matched. So then the challenge is how to expose the current completion options from a zsh prompt. The only prior art I could find was this hack on github: https://github.com/Valodim/zsh-capture-completion But that's not particularly useful to me. I could imagine two high-level ways of doing it: - A "push" strategy where after every keystroke the completion code is updated, a la https://github.com/marlonrichert/zsh-autocomplete. Only in this case the results would be written to a file named after the WINDOWID (this is all under X11). After that, watching for file updates and tracking the focused window and notifying talon accordingly is easy. - A "pull" strategy where some callable function or widget can return the results of a completion (without altering the prompt of course). That function/widget could be triggered by trapping a signal or by reading a packet from a socket or something. After that, I would be able to hook the rest of the system together I believe. I'm pretty new to trying to hack zsh. I spent a bunch of time trying to read docs and make things work but I still don't really have any idea of what I want is possible (or how much coding it would take to become possible). Does anybody have any ideas or pointers for how to do this sort of thing?
On Wed, Dec 30, 2020 at 2:27 PM Ryan Beethe <ryan@splintermail.com> wrote: > > I am trying to set up voice control software (talon with a wav2letter > backend) to work well with shell completion. I would like to be able to > capture what the results of a completion *would* be if it ran, but > without the normal showing of results. I suspect that Roman's zsh4humans configuration that passes completion results to fzf could be modified to pass them to the speech backend instead. https://github.com/romkatv/zsh4humans/ Specifically https://github.com/romkatv/zsh4humans/blob/v5/fn/z4h-fzf-complete