Need to pass arbitrary mode name as a parameter to HA


I have an integration for HA which helps me to do the set up of the lights(for example) using voice interface. In order to appeal to the devices I made them declared under ids: device_1, device_2 and so on… And I have a descriptor with the text which contains the human friendly name. So for example, when I say: “what devices(or lights) are in the livingroom”, the integration finds all the diveces in the room, reads their ‘slot names’, which rhasspy recognises, reads human friendly name and sends to TTS merged message: “there are found following lights in the livingroom: device 1 - lights by the TV, device 2 - lights at the top…”. So now I know the corespondance of the device_id, mode_id and their normal names and can say: “set device 2 to mode 1”. This is pretty convenient for the intent recognition and when you use it, because you will never remember the device names and mode names anyway when you have many rooms, devices and preset modes for those…

The problem: I need to pass an arbitrary text, obtained from STT rhasspy does by means of Kaldi, to HA as a parameter in the event. I have tried everything that came to my mind, but nothing was successfull. What I expected to be happening is that I have an intent
((set the name for the ($entity):{target}): )

and some text spoken, that to be embeded to the event data.
For example, when I say: " set the name for device1 my favorite lamp that I bought in Venice " I get the parameter ‘target:device1’ and embeded text “my favorite lamp that I bought in Venice”. So now I can use this text to put into a descriptor field and my system would know, that device1 has this name.

But at the moment, no metter what I do I only get “set the name for the device1” in the embeded text, while everything that is said beyond the text in the sentences.ini is ignored.

I have tried a mixed-language model with weight 0.05, but it did not have any effect.

Does enyone know what can be done to make ‘untrained’ part of the STT be saved into raw text data? Or any other ideas on how to implement this?