I'm trying to create a custom slot type for client name lookup. Some which have very custom words, such as people's last names or company names.
These vary majority of them are not common english last names. In order for Alexa to interpret the correct phrases I have to load all the variations in slot values. This can be a lot as each user in our dataset has many clients.
I've tried loading all the slot value variations though this exceeds the 1.5MB interaction model JSON file.
Even if there isn't this limit and I can load all the slot values, the JSON file would be huge (500MB easy). Likely a fairly bad design.
The other option is I can try to use AWS transcribe or my own natural language model to interpret and turn the speech into text. However Alexa does not give access to the users voice input file.
Is there anything you guys recommend in terms of design in order to create my own custom slot type or connect it to another aws service to solve this problem?
Is there a way to create a custom slot type and link it to my skill such as the built in
AMAZON.[SlotType] that doesn't impact the model JSON file size?
Any recommendations is appreciated. Thanks.