The boffins over in the Fraunhofer-Gesellschaft institute got a lot of press with their printable batteries last week, but buried a little deeper is something possibly even more interesting, a search engine for television programs. Existing voice-to-text transcribers rely on a list of words that can be understood by the software. This list can only contain a limited number of words and it needs to be maintained and updated with new terms like “financial crisis”, “Michael Jackson” or “Obama” in order for it to remain relevant.
The Fraunhofer Institute for Intelligent Analysis has taken a different approach and have developed a speech recognition program based on syllables, thereby reducing the number of terms the software needs to learn to a more manageable 10,000 entries. When the software analyses a video, the speech is reduced to terms such as fi nan cial cri sis which a speech algorithm then converts into text. The text can then be searched via a conventional search algorithm and 10,000 hours of pre-recorded broadcasts can be scanned in milliseconds.
The Institute claim that 85% of spoken words can be recovered and 99% of located contributions are correct. It also means that each broadcast only has to be indexed once, unlike the existing method which needs to be redone with new keywords. Fraunhofer is licensing the program and making it available for use. [Photo from Flickr]
The journalist recalls more or less what Ulla Schmidt said regarding the health reform, but needs the exact wording to be able to cite her. A new speech recognition system helps to search TV broadcasts. It does not need to be updated and so does not entail any running costs.