I still feel on device Dication is not accurate yet.

Coming here as Developer Relations at AssemblyAI, specially where entities need to be maintained such as names from different cultures.

Can we try comparison by switching API call to our Dictation API.