question

I wish Amazon would add this to their FAQ... Alexa was not designed for general speech-to-text. It was designed to work with a restricted vocabulary defined by your utterances. You can trick Alexa into letting you create a "generic slot" using what you've done. But the quality of recognition will be poor. Unless users are particularly simplistic in their naming, you will almost certainly not get what is on their list. So you are going to have to do some sort of phonetic comparison (e.g. DoubleMetaphone) to compare what Alexa thought the user said, to the possibilities of the user might have said. As to your other question, no, you can't dynamically update your interaction model. It's been requested a number of times. You can go +1 any of the feature request threads asking for this. But I wouldn't build a business plan on it. ... Now, one way to work the problem is this. Don't let them use free form naming in your mobile app. If the name is purely abstract, you could pick, say, eight colors and eight animals, and let the user choose a memorable combination of the two (e.g. Purple Capybara). If it is, say, something like home automation, you can do the same, but make the first one a room type (living room, dining room, etc) and the second a function type (lights, temperature, etc). That will be more natural if that's what your functionality is. Back in your Alexa app, you now no longer need free form speech recognition. You can create one custom slot with the list of options for the first word, and a second slot for the options for the second word. Now you are back in the territory Alexa is good at. Yes, it isn't ideal for the users. But it might be the most usable choice if you want to go this route.

Thanks for the feedback. We have tested the custom slot option and it works well...it just isn't the ideal solution for us, although it may be the best one at the current time. I will definitely +1 the idea for being to dynamically update our interaction model.