Speech Recognition

AbstractThe ease at which humans use speech understates the complexity of speech recognition for machines. The fundamental difficulty with speech recognition is the overwhelming variation in the production of speech. An automatic speech recognition system must successfully address this variability in speech before any useful output can be provided to the user. Typical automatic speech recognition systems use sophisticated mathematical models, Hidden Markov Models, to build a catalogue of speech features. These models, along with models of language grammar and structure, are used to recognize an incoming speech signal into its most likely word sequence.Today, speech recognition systems are increasing deployed in areas where automation provides both convenience and cost savings. Key problems such as the ability to recognize speech in a variety of environments and over different speaking styles prevent the widespread use of speech recognition technology. In the future, solutions to these problems and others will pave the path for many new and exciting uses and applications of speech recognition.