What's the high level approach required to build something like this yourself? Does it involve relying on speech to text with timestamps and then a series of c

What's the high level approach required to build something like this yourself? Does it involve relying on speech to text with timestamps and then a series of cuts based on that?