Wednesday, July 9, 2003

Confused about speech technology:The technical kinks, high costs and application misfires that have held back the acceptance of speech recognition and activation are being ironed out. [...] Now the directions of both research and marketing have changed. Rather than developing a machine that can converse, researchers are creating computers that can understand speech as a function of probability, the basis of much of Microsoft's artificial intelligence work. [CNET News.com] Huh? Every significant speech recognizer for the last 20 years has been based in probabilistic models. Much of that pioneered by IBM Research. Well before there was a Microsoft Research. Faster machines, more extensive data collection, better mathematical models, and better algorithms have been steadily improving speech recognition. Microsoft's general work on probabilistic modeling and machine learning is excellent, but much of the specific work in speech recognition has been going on elsewhere. Current techniques are still brittle and limited in scope. Predictions like the ones in this article were made ten years ago, too. 9:01:22 PM