Hi,
AFAIK, There is no way to do threshold signatures non-interactively for the general case of t out of n. Even if you are willing to maintain additional data structure on top of the standard and change verification algorithm (see for example appendix B in [1] where they use bitmaps).
The best way that I came up with so far (which I plan to implement in the library) is to take SS01 paper [2], this also the paper cited in bip-schnorr [3], and to replace Pedersen VSS with Feldman VSS (Feldman VSS implementation can be found in [4] ). Basically taking the DKG from GG18 without paillier and the dlog pok (threshold ecdsa paper [5]) and use it for the threshold schnorr DKG and for the ephemeral key distributed generation. This will cause the lost of Robustness but will be more efficient.
Generally speaking - the purpose of using threshold security is to replace hw security. The assumption is that you would rather trust that no more than t out of n different machines will get corrupted at same time than to trust one secure hardware. Maybe that relax a bit the demand for using air gapped devices?