Not an expert but ...

RFC2833 is a specification for synthesizing the tones.
This method should be less distorted at the far end.
This method is likely source of discussion of spurious tones from certain higher pitched voices.

Inband combines the tones with the voice channel.
I do not know if this is ever needed because tones don't get generated at the far end?

Perhaps automatic can determine that RFC2833 is not working and downgrade to Inband?