Speech recognition

Article By:

Juang, Biing Hwang Lucent Technologies, Bell Laboratories, Murray Hill, New Jersey.

Last reviewed:June 2020

DOI:https://doi.org/10.1036/1097-8542.643100

Speech recognition, in a strict sense, is the process of electronically converting a speech waveform (as the acoustic realization of a linguistic expression) into words (as a best-decoded sequence of linguistic units). At times it can be generalized to the process of extracting a linguistic notion from a sequence of sounds, that is, an acoustic event, which may encompass linguistically relevant components, such as words or phrases, as well as irrelevant components, such as ambient noise, extraneous or partial words in an utterance, and so on. Applications of speech recognition include an automatic typewriter that responds to voice, voice-controlled access to information services (such as news and messages), and automated commercial transactions (for example, price inquiry or merchandise order by telephone), to name a few. Sometimes, the concept of speech recognition may include “speech understanding,” because the use of a speech recognizer often involves understanding the intended message expressed in the spoken words. Currently, such an understanding process can be performed only in an extremely limited sense, often for the purpose of initiating a particular service action among a few choices. For example, a caller's input utterance “I'd like to borrow money to buy a car” to an automatic call-routing system of a bank would connect the caller to the bank's loan department.

The content above is only an excerpt.