The Lecture Browser is a web interface to video recordings of lectures and seminars that have been indexed using automatic speech recognition technology. You can search for topics, much like a regular web search engine.
Follow the link to the Lecture Browser and you can experience this search adapted onto the lectures of MIT. I suppose from here it will not take too long and in your search engine, in addition to text search and image search, you will have audio search.
As a side note on image search: for now, the search is with keywords in the text that accompany the pictures, but technology is also being developed to recognize text inside the picture (either captions or captured text) and include that in the search.