Add more caching

[adriansteffan]

As we build onto this project, it might make sense to add some more caching steps to store intermediate inference results (e.g. save speech transcriptions after each file)

(In a similar vein: The console output could use a bit of cleanup to more clearly communicate what step is being worked on at the moment, e.g. some models being silently downloaded within the transcription function might lead to unclear interpretations on why the scripts isn't progressing)