Command recognition algorithms?

Question

Speech recognition

Hello.
The task is the following:

1. The user pronounces the command script turns it into text. This is nonsense, you can use external services, that's not the point
   
2. then the program must understand what is what. in command
   For example:
   
   • turn on the lamp
     
   • white light turn on
     
   • turn on the light in the corner
     
   • turn on two lights
     
   
   etc. The system must understand that:
   
   • "enable" is an action
     
   • "lamp" - subject
     
   • "white lamp" - subject
     
   • "lamp in the corner" - subject
     
   • "two" - the number of items
     
   
   

The example uses "Lamp", but I don't know what items can be. We need something universal. The principle roughly resembles the analysis of a sentence into sentence members, predicate, subject, definition.
It turns out that at the input we have the text of the command, and at the output: the command in the form of an object, quantity, action.
In which direction to dig?

Answer 1

For processing short commands, there is no point in using neural networks and other heavy artillery.
The received commands (in the sense of action) are known in advance, all the quantitative parameters and the objects themselves too. Then the processing of the command is reduced to the processing of word tokens to determine the action, objects and their characteristics, and all of them are known to the program at the time of receipt.

Example

Известные действия: {включить, выключить}
Известные объекты с характеристиками: {светильник={синий, в углу}, светильник={белый, на тумбе}}
Поступающая комманда: "белый светильник включить"
Токены комманды: {белый, светильник, включить}
Шаги алгоритма:
1. Находим токен-действие - "включить";
2. Находим токен-объект по ключу известных объектов: "светильник";
3. Оставшиеся токены ("белый") - характеристики. Проверяем все объекты с ключом "светильник" на наличие характеристики "белый", находим нужный;
4. ???;
5. Комманда обработана - найдено действие и объект.

Answer 2

1. Definition of the part of speech: https://habrahabr.ru/post/152389/
2. To compare the received text (from the microphone) and pre-created expected commands (or their templates) - fuzzy search Stumper API .

Answer 3

Yandex has a whole stack of technologies for solving problems of speech and text recognition in Russian. Look in this direction, for example - https://tech.yandex.ru/mystem/doc/usage-examples-d...