Overview of speech recognition

Speech recognition is a complex task as it has to consider several sources of variation in the data. Some of these are variations in the speaker, the size of the vocabulary, ambient noise, accent, speaker characteristics, and so on. For example, one person would be saying a word like apple very fast compared to another who might be saying it more slowly like app........le. In both cases, the speech recognition system should produce the word apple. The most common approach to speech recognition uses Hidden Markov Models (HMMs) since speech data can be considered as a stochastic or probabilistic process, and for a short time, slices can be considered to be independent of time. The HMM model can be trained on short slices of the speech data that represents a phoneme or word. It can then predict the next phoneme or word that can be combined together...