Best approach with similar words

I’m trying to differentiate between similar words
Google fails on this in as much as ok doogal
triggers it.
However I am going for a smaller user base, just me.
I want to differentiate between terms (in French) like
mon (my) and non (no)
sinon and Simone (the n is stressed in Simone)

Any ideas???

Hi @greg_dickson It’s a though task to do this in a generic way, but if it’s only you then probably doable. I’d just collect some samples and see if the model reaches any accuracy, if so, you can expand the model.

Yeah thanks @janjongboom
I spent the day tweaking the parameters all over the place getting lost many times. But what I came up with after final deployment was pretty good. After testing the real world deployed version it already works as good a doogal :slight_smile: ,
With only 600 samples in each group so a larger base will definitely help.
But some of the results were the opposite of what I expected but black magic is often like that.
It is almost one needs an AI to refine the AI,
But thanks for the feedback.
This rabbit hole has certain become a lot less deep since I started to use EdgeImpulse.

1 Like