Comment by MichealCodes

I don't think we've had the transformer moment for audio training yet, but yes, in theory audio-first models will be much more capable.