I doubt audio is transcribed on each device rather than once in the cloud, but imagine for a moment... what if it was? Or if Amazon gave you access to a server to redo the transcription work in real time for each user? Would Amazon would be distributing the capability to create a derivative work rather than (what is arguably) a derivative work?

Should Apple be prevented from distributing Logic Pro audio software because it could be theoretically used to create unauthorized derivative works from music downloaded from iTunes? Should Google/Alphabet be prevented from serving search results for "youtube-dl" because it is the distributor of copyrighted music videos? Should MEGA.nz be prevented from sending data whose copyright status it has no ability to determine by design? At what point does responsibility pass to the person clicking the "transcribe" or "remix" button?

(IANAL)