A user’s voice can be used to authenticate them in any scenario where audio can be recorded and processed. A text-dependent (active) voice algorithm requires the user to repeat a short audio phrase or voiceprint. Because the spoken phrase itself is a constant, the voice biometric matching is extremely accurate and in many cases, requires no additional authentication for most users. A text-independent (passive) approach allows
The more they speak, the more confidence the passive voice algorithm has that it’s really the person doing the talking. For this reason, passive voice biometrics are extremely well suited to lengthier, more conversational situations like interactions with a live contact center agent.
How does voice authentication work?
Voice biometrics use a trained background model constructed from thousands of utterances to represent average speaker characteristics. It is important to use representative audio from people with similar demographics, devices and capture environments to the target population. During verification, the captured audio is compared with both the background model and the user’s own enrollment data to determine if there is a valid match.
Voice verification checks that the verification utterance originates from the same person who recorded the earlier enrollment data. In text-dependent speech a specialized speech recognition algorithm is often used to confirm that the correct expected phrase-words were uttered.
Voice recognition is best suited to authenticating a user, although it can also be used for identification in conjunction with a forensic voice expert. The accuracy of voice biometrics can be affected by the microphone used, presence of loud background noise, and the quality of the communications channel the audio is transmitted over. Voice authentication results are best when a person uses the same capture microphone each time, i.e. their personal smartphone..