Today, many different types of algorithms can be used to interpret human speech. An important feature of many of these algorithms is that they can learn to recognise the sounds of different voices and accents. In ideal conditions many algorithms can recognise more than 90% of speech. When situations are less than ideal—for example, in a noisy environment or when a speaker has an unusual dialect—the recognition might be severely impeded. In a control room situation, one can expect the use of a rather limited vocabulary, for example, in emergency situations. With a limited vocabulary the speech algorithm can be developed to recognise many different accents, even when the background environment is noisy and disruptive.
A speech generation system is the mirror image of a recognition system. It consists of a device to generate messages in the form of symbol strings, a speech generation algorithm to convert the symbol strings to an acoustic imitation of human speech, and a human listener. A speech generation system operates within the context of the user’s working environment: