Comparing transcriptions

[Joshfindit]

This is quite a minor one, but it might inspire people to create more powerful tools:

To compare the transcription between models I was able to get meaningful results by:

1. Transcribe the audio with each model that's to be compared
2. Take the txt file and change it to be 1 line per "word" (space as the delimiter)
3. Run a diff on the two files

Output will be like:

monday
sounds
good
thank
you
all
right
thanks
a
lot
so
Oh
You

Monday
Sounds
good
Thank
you
Alright
thanks
a
lot
Cheers
Cheers
Bye
bye
Thank
you

If doing a simpler compare, all text can be normalized to lower case before comparison.

This gives a nice readable diff output that can be scanned through where gaps are immediately obvious as are words that are transcribed differently for each model.

If someone wants to lead the project, I'd help build a tool that runs multiple models and lets the user choose the best option for each word/sentence, for those use cases where correctness is more important.