Skip to content

Latest commit

 

History

History
48 lines (33 loc) · 2.88 KB

TODO.md

File metadata and controls

48 lines (33 loc) · 2.88 KB
Raw

General

  • "turn on the wifi, the airplane mode and the flashlight" --> ????? There are no commas on speech recognizers and there's no "and" in the place of the comma, but someone could say it like that and then oops....

  • "stop and play the video" - with the repeated commands check enabled on the verification function, this won't work... See if it can be linked with the {TURN_OFF, something} idea. Though, "turn off... turn on the wifi"... The "and" has an important role on this issue. Without it, might be a correction of what was just said (like the last example of the Wi-Fi).

wordsVerificationFunction()

  • Put the assistant calculating some parameters for the function (by assistant I mean the submodule itself - or Android or whatever is using the library, if necessary, though more parameters will be necessary to add and the types are restricted by Gomobile).

For example, it could see if with "set an alarm", "set" appears only in the first sub-slice, it would set the parameter to exclude that word from the list, but not if words are repeated on the sub-slices! Infinite parameters on the function but some automatic! Adaptable assistant by itself...!

Based on the content of some function slices, it would adjust other parameters automatically without needing to go there manually and hardcode it.

NLP

  • Also, ONLY USE NLP WHEN IT IS REALLY NEEDED! If not REALLY needed, use the function without its help (it may fail, like with "turn on the wifi", on which it thinks "turn" is a name and not a verb - works in a bigger sentence though).

  • Use NLP to exclude other adjectives from the words_list if the only one was said already, for example? (Or a verb or whatever) Of course, works only (as said) when there is only and only 1 adjective to form the command.

Example: "take a frontal and rear picture", which becomes "take a frontal take rear picture" - because picture is a name, which is not included with the "and" replace function. "take a frontal picture and a rear one" doesn't work either. ---> It needs to know it's taking a picture here to then be informed of its type! <--- What I just wrote is a core idea of the supposed implementation!!! Currently, for that issue to work, it must be said "take a frontal picture and a rear picture". Also, "rear one", rear is an adjective, so "one" refers to that adjective, which is referring to "picture", hence "one" refers to "picture" - try to use this.

NLP + wordsVerificationFunction()

  • Re-implement the verification function to also use NLP: "turn on the wifi" --> "verb [no idea] name" or "verb name [no idea]". Then it checks if the words on the 'words_list' are in the intervals of words that it found. For example, a verb is at index 12. Then from index 13 until the next name, there must be an "on". Then from the next name to the next non-name (since we're looking for a name), "wifi" must be there. And so on.