The goal of speech-enabled applications has always been to allow callers to obtain information and perform transactions simply by speaking naturally. Recognition of free-form conversation is not yet a reality, and speech recognition has sometimes been over-hyped in the past. However, speech-recognition technology is now proving itself commercially viable in a number of customer service applications. Every day, thousands of callers are using their own voice to perform millions of business transactions.

Since the late 1980s, the U.S. government's Advanced Research Projects Agency has supported significant research in computer speech recognition and understanding systems. This research has improved recognition accuracy and increased capacity to recognize larger vocabularies.

The earliest speech-recognition engines provided only discrete number recognition and did not understand a caller's speech unless each word was separately and distinctly enunciated. Improvements in the recognition of continuous speech—words spoken in an unbroken string—now allow users to talk normally without artificial pauses between each word. Continuous speech recognition requires greater processing capacity and may require artificial intelligence (AI) or other associated technologies to ensure a high degree of accuracy.

Early speaker-dependent dictation systems had to be trained to understand the speech of one specific user. Now, speaker-independent recognition technologies allow IVR systems to interpret the speech of many users. Today's speech-enabled IVR applications serve large numbers of unknown callers without prior training of the system.