Automatic Speech Recognition — Plug-In MAP, Kernel Methods, and Hybrid Systems

This note gives a short introduction to automatic speech recognition. Various approaches are discussed, including the well established plug-in MAP rule with hidden Markov models, the more recent methods known as kernel methods, and hybrid systems that combine hidden Markov models with kernel methods. The note also includes some new and promising methods for future research. Section 1 concerns the problem of pattern recognition. In Sec. 2, automatic speech recognition (ASR) is defined as a particular case of pattern recognition and various approaches to ASR are discussed.