For instance, I searched for the word friendship and got a result titled “Obama Anniversary of Israeli Independence”. Moving the time bar line towards one of the yellow markers, I’m hearing Obama say, “... let us renew the friendship between our nations ...”. Another search for bubble sort returns a quip from a Q&A Obama held at the Google headquarters, where Obama is answering a programming question.
While the hits returned look precisely correct, hovering over the yellow marker pops up a snippet which shows Google doesn’t get it completely right... like when they transcribe a piece as “... to recognize Israel and are fighting for friendship wolf they ...”. Google admits as much in their blog post on this service, saying snippets “may not be 100% accurate”. They add that speech recognition “is a difficult problem that hasn’t yet been completely solved”.
This gadget shows that using speech-to-text technology, even when it’s very limited in scope right now, is a highly useful feature for video search. If and when Google ever rolls this out for all video content they indexed (or perhaps just for their YouTube content for starters), it could become an extremely interesting research tool. Besides just aiding search, extracting textual content from videos could also be used by Google to ...:
Any of these features could have two effects: first, it could make Google Video or YouTube a more attractive place to search for and watch videos, and second, if a feature is only rolled out for YouTube (and if a content creator would like to have this feature for their video) it could make YouTube a more attractive place to upload videos.
>> More posts