|
Speaker biometrics are often referred to
as speaker recognition in the speech development community.
Speaker recognition is in contrast to speech recognition which recognizes
the content of what the speaker says. As described on the
biometrics basics library page, speaker verification verifies the
speakers claimed identity and speaker identification identifies a speaker
from a set of known speakers.
There are THREE TYPES of speaker authentication;
-
TEXT DEPENDENT speaker authentication is dependent on a specific
word or phrase used during verification. The speaker has enrolled
with the same or very similar words. Text dependent methods are
typically more accurate given limited amounts of speech.
-
TEXT INDEPENDENT speaker authentication is independent of the
words used during verification. The speaker has enrolled with free
form controlled speech. Text independent methods are more flexible
than text dependent methods. It can be used simultaneously with speech
recognition to conveniently recognize what a person says as well as who they
are.
-
TEXT PROMPTED or challenge-response prompts the speaker for randomly
selected words or phrases to be used for verification. Text prompted
methods protect against playback of recorded speech.
|
