Add native audio transcription using whisper for speech detection

[vladtaylor]

for speech detection objects, it would be nice to for frigate to "guess" what the speech actually said.

Can be handy to search for clips that have speech and include certain keyphrases.

I have a rough POC working on my personal github repo called frigate-listenerr, it kinda works, but not terribly well, hoping to do some more testing and improve some fine tuning.

I have it add it to the description of the clip object and stored locally.

Lets say I have some troublesome neighbours, that have "vivid" conversations that often turn into events which require intervention from the authorities. Having something keep track of what was said, may prove useful one day.

[vladtaylor]

https://github.com/vladtaylor/frigate-listenerr

[NickM-27]

This is already implemented for 0.17