Analyse video/audio file of someone reading text and add bouncing ball

Our product features 1000's of books used by schools and kids learning to read. Our students watch video of someone reading the book, see example below: [login to view URL] We want to add a 'bouncing ball' visual feature to all our videos, see this example [login to view URL] We're ideally looking to automate the feature, but happy to explore our options to cost effectively achieve our objectives. Option 1 - process the audio/video & produce a file of the words spoken and the time the person spoke them (so we can use that on our platform to add the bouncing ball) Option 2 - integrate to some of the new natural language processing API's, Machine Intelligence API's, etc. What new tech or services could we use and should investigate Option 3 - manually create the data needed (we'd recruit low cost resources so probably not a solution for this project, but maybe there is a mix we should consider of automated and manual?