Speech Recognition Moves from Software to Hardware

Speech recognition has long promised a natural way to improve user interaction with computers, cars, and other devices. During the past 30 years, researchers have gradually upgraded the technology to the point that it is used in a number of these settings. However, because of limitations in processing power and other factors, the applications typically have been relatively simple, and speech recognition has not been widely used, despite the growing desire to implement it in PCs, cell phones, applications that automate home utilities and entertainment devices, and other systems. Researchers have been working on implementing speech recognition in dedicated processors for about 20 years, but the chips still have limited capabilities and work with only relatively small vocabularies. As such, few companies sell speech chips. Now, though, scientists are interested in developing high-end speech chips that work with large vocabularies of words and that recognize continuous speech. Despite its promise, speech-chip technology faces technical and marketplace challenges