I was listening to lots of podcasts and audiobooks while insulating and decorating a bedroom.
I noticed there were occasionally repeated sections left in where the voice artist wanted to re-record. Also I was constantly having to turn the volume up and down between podcast episodes, which was tricky when your hands are covered in glue and paint.
I did some research and found there were audio standards for things like loudness (which not many podcasts were following). I figured I could detect restarted sentences by first performing speech recognition on the audio.