Difficulties
Be aware of this technical and ethical pitfalls.

Privacy Concerns
Recordings of biometrical data, like a person's voice, create privacy concerns. 
My goal was, to pass audio through the tool chain without the need for a recording that exists as a file on the device. 
This turned out not to be feasible with whisper-stream's current architecture.
 Even adding the automatic removal of recording files to the 
meeting-stop script, doesn't change the reality, that TranscriptOMatic works — in its current state —, only by first recording the audio to a file on the device's storage.
Keep this in mind when using your own variant of TranscriptOMatic. Communicate the circumstance clearly to enable informed decisions of all participants. 

Voice Activity Detection
In the whisper.cpp built used in this example, Voice Activity Detection does not exist. 
mela@Cox:~/meetings/lib $ error: unknown argument: --vad
Voice Activity Detection, if present, enables whisper.cpp to only transcribe when somebody is actually talking, instead of also acting on silence or background noise. This preserves computing resources and reduces hallucinations while nobody is speaking.