Breakthrough in computer speech-recognition.
The Internet may not be making us smarter, but it may be getting smarter about us. Recent breakthroughs in speechrecognition technology point toward a future where Web crawlers recognize more of the words we speak.
Researchers at the Fraunhofer Institute for Intelligent Analysis have developed a type of search engine that works for spoken words in television broadcasts. Other similar speech-recognition programs attempt to match what the system hears to words stored in a digital database. The Fraunhofer program recognizes syllables and pieces together spoken words based on their small parts, allowing for almost 99% accuracy, the researchers claim.
“Our system is based on a syllable thesaurus instead of a word thesaurus,” according to researcher Daniel Schneider. “Conventional speech recognizers can only discern a limited number of words, while the total number of words in existence is too vast to handle. The number of existing syllables, on the other hand, is manageable. With about 10,000 stored syllables, we can make up any word.”
The program is also able to differentiate between speakers and scan thousands of hours of broadcasts in just a few milliseconds, the researchers assert. Users can look for bits of spoken dialogue based on when comments were made, what was said, where, or by whom.
Students, detectives, or snoops could also use the program to analyze surveillance footage (provided people’s words are clear enough to be heard on the recording). As more people spend more time under the lenses of cameras and in the presence of microphones, and as more footage from those cameras and devices goes online, some interesting if not troubling implications for privacy emerge. A spoken-word search engine could one day theoretically allow someone with little more than a smart phone to look up any recorded conversation between any two people that’s occurred virtually anywhere a microphone was present. -Patrick Tucker
Source: Fraunhofer-Gesellschaft, http://www.fraunhoher.de.
Originally published in THE FUTURIST, January-February 2010