Know the Difference Between Recognition and Verification

One question speech system users need to answer before they actually make a purchase is “do I really know what I want?” Often, newcomers to the industry are confused about the difference between speech recognition and speaker verification and have expectations which are not impossible, but just inappropriate for the system they are investigating. Hopefully, the following will be helpful. “Speech recognition” refers to the ability to recognize what words have been spoken. A computer is programmed to take specific actions, according to the words recognized, the same as it is programmed to respond to specific keyboard commands. “Speaker verification” on the other hand, refers to the ability to recognize who is speaking various words. The computer is programmed to take certain actions if the persons speaking is who he or she claims to be, the same as a correctly entered keyboard password enables access to specific computer functions. Even if the words spoken into the system are exactly the same (for example, “ten”, “strawberry” or “ninety-six”), they will be handled differently by the different types of systems. With recognition, the computer interprets the words as commands to do something. With verification, the computers uses a speaker’s “voiceprint” to determine if the speaker is valid, or an impostor. All verifiers follow essentially the same sequence of operation. First, a new user “enrolls” his or her voice by speaking certain phrases in response to screen prompts or aural prompts. A unique voiceprint is created from the responses and stored for use later. Each voiceprint is identified by a unique account number, either selected by the user or assigned by the system and given to the user. Subsequent system operation starts with account number entry, to retrieve the reference voiceprint. The user’s voiceprint is then compared against the reference voiceprint and the system determines whether the voiceprints match. Some verifiers incorporate recognition as a supporting function. Also, speech recognition can be used to defeat the use of tape recordings of a person’s speech. While there are certainly other means of verifying a person’s identification, speaker verification has an advantage over them all. Voices can not be stolen, borrowed, copied, or guessed, as with PINs and passwords.