Machine Speech Recognition

Speech represents the most natural means of human communication. Machine speech recognition, often called automatic speech recognition, is the automatic process performed by machine or computer to transform a speech utterance into a text consisting of a string of words. The term machine aims at making the distinction between machine speech recognition and human speech recognition (human speech perception). Machine speech recognition is also different from machine speaker recognition, which is the automatic process performed by machine to identify a speaker or to verify the identity of a speaker based on his or her voice. After a brief overview of the general steps involved in machine speech recognition, this entry introduces two of the most prominent machine speech recognition techniques.

There are different techniques for machine speech recognition, but generally speaking, they consist of a common series of steps. First, the acoustic waves of pressure corresponding to the speech utterances are transformed into electric signals by a microphone. These electric signals are then transformed into a string of feature vectors, usually called acoustic feature vectors. The feature vectors are representations of the spectrum and energy of the speech signal over short periods. Then, the extracted string of acoustic feature vectors is matched against previously stored models of sentences, words, syllables, or phonemes. The text string of words that best matches the incoming string of acoustic feature vectors is presented at the output of the machine speech recognition. Based on the type of input utterances, machine speech recognition can be classified as isolated word recognition or continuous speech recognition. Based on the generality of the models, machine speech recognition can be classified as speaker dependent or speaker independent. Based on the size of the vocabulary, machine speech recognition can be classified as small vocabulary (up to 100 words), medium vocabulary (up to 1,000 words), or large vocabulary (up to hundreds of thousands of words). Applications of machine speech recognition include voice dialing (e.g., digit recognition), command and control, form filling (e.g., data entry), web search by voice, and dictation (e.g., speech-to-text word processing).