|
FAQ
- What is voice verification? How is it different than voice biometrics?
- What is a voiceprint?
- How much storage do the voiceprints require?
- What are some other common terms for voice verification?
- What is the difference between speaker verification
and speaker identification?
- What is the difference between text dependent and
text independent voice verification?
- How is voice verification different from speech
recognition?
- How secure and accurate is voice biometrics?
- Will a cold affect the accuracy of voice biometrics?
- What about voice biometrics and privacy?
- What is voice verification? How is it different than voice biometrics?
It is common knowledge that each person can be identified by the unique
characteristics of his or her fingerprints. As with fingerprints,
it has been proven that each person can also be identified by the
unique characteristics of his or her voice. Voice verification is
the process of comparing a voice sample with a stored, digital voice
model, or voiceprint, for the purposes of establishing or verifying
identity. Voice verification is a type of voice biometrics, but voice
biometrics can also include other techniques, such as one-to-many or many-to-many
voiceprint matches.
top
- What is a voiceprint?
A voiceprint is a digital representation of some of the unique characteristics
of a caller’s voice. A voiceprint includes measurement of physiological
characteristics, such as the nasal passages and vocal chords, as well
as the frequency, cadence and duration of the vocal pattern. A voiceprint
is not a recording or sound file – it cannot be played back
into a voice biometrics system by an imposter. The voiceprint is a voice biometric identifier.
top
- How much storage do the voiceprints
require?
Voiceprints require, on average, approximately 80Kb of storage.
top
- What are some other common
terms for voice verification?
Voice verification is sometimes referred to as voice authentication,
voice biometrics or speaker verification. It is sometimes incorrectly
referred to as speaker identification. Voice verification and speaker
identification are both types of voice biometrics, but there is an important
distinction between the two, as described below.
top
- What is the difference between
speaker verification and speaker identification?
In voice verification, a subject’s identity is confirmed by
matching his or her voice against a single voiceprint. This is typically
done when there is some information provided to indicate the subject’s
identity, such as an account number, and the voiceprint matching is
done for confirmation. Spike Server operates in voice verification
mode.
In speaker identification, the subject’s voiceprint is matched
against several voiceprints. This is typically done when no other
identifying information is available, such as in the case of surveillance
audio analysis or when a caller has forgotten his or her password
or account number. Both speaker verification and speaker identification are types
of voice biometrics.
top
- What is the difference between
text dependent and text independent voice verification?
The majority of voice verification systems are text dependent, meaning
the user must state a specific word, phrase or digit sequence in order
to be verified. This is ideal for combining knowledge verification
(password, account number) with biometrics. There are some voice biometrics
systems that are text independent, meaning a speaker can be verified
based on any voice sample. This type of system is most suitable for
systems that run in the background, such as for surveillance operations,
where it is not possible to prompt the subject to state a specific
word or phrase.
top
- How is voice verification
different from speech recognition?
Speech recognition, sometimes referred to as voice recognition or
automated speech recognition (ASR), is a technology that analyzes
a voice sample to determine what was said, while voice biometrics
analyzes a voice sample to determine who said it. In other words,
ASR recognizes words, while voice biometrics recognizes identities.
top
- How secure and accurate is
voice biometrics?
When applied properly, voice biometrics can help organizations improve
security dramatically over current knowledge-based methods of verification,
such as password, PIN or challenge questions. In contrast to a knowledge-based
system that simply checks for the right password or PIN, voice biometrics systems
actually verify that the authorized user is the one entering the password.
The accuracy of the voice biometrics system is affected by several factors;
including the duration and number of the voice samples provided during enrollment
and authentication, and background noise levels.
Accuracy is typically measured by two variables: False acceptance
rate (FAR) and false rejection rate (FRR). A false acceptance means
that an imposter has made it through the system. A false rejection
means that a valid user has been rejected. With Spike Server, security
thresholds can be adjusted to prevent virtually all false acceptances.
Setting the security threshold to “High” will decrease
the FAR, but will result in an increased FRR, and vice versa.
Essentially, the voice biometrics system can be appropriately adjusted for
various security requirements. For example, conducting a million dollar
stock trade may require several voice samples and a very low FAR, while a
small transfer from a savings account to a checking account may only require
a brief voice sample at a lower security threshold. The point at which the FAR
is equal to FRR is called the Equal Error Rate (ERR).
Adding knowledge verification (passphrase or PIN), in addition
to voice biometrics, adds an additional layer of security. Knowledge
verification can easily be accomplished with Spike Server’s
built-in automated speech recognition capability. Voice biometrics
can either displace or augment current knowledge-based authentication
systems.
top
- Will a cold affect the accuracy of voice biometrics?
Not all of the parts that make up your voiceprint are affected by
a cold; the voice verification engine can still verify a caller that
has a common cold. Serious throat illnesses such as laryngitis may
negatively affect system performance.
top
- What about voice biometrics and
privacy?
Like any technology, biometrics is defined by its usage. Biometrics
is neither inherently privacy-protective nor privacy-invasive. The
uses and controls of the biometrics system are what impacts privacy.
Take, for example, the case of a bank using voice biometrics to identify
account holders prior to allowing a credit-card activation. The bank
is protecting the client from identity theft, and is therefore protecting
privacy.
Some people are concerned that a voiceprint could be compromised or
shared, resulting in a threat to privacy. The caller’s voiceprint,
if somehow stolen from the server, cannot be used to access a voice
biometrics system. Voice biometrics systems like Spike Server require
the caller to provide a short speech sample during authentication.
Even though a voiceprint is derived from a speech sample during enrollment,
this speech sample cannot be obtained by reverse-engineering the voiceprint.
In other words, a stolen voiceprint is essentially useless.
Any personal data used in Spike Server, such as account numbers or
recordings, are encrypted. Diaphonics also employs proprietary anti-spoofing
techniques to prevent the use of recordings designed to fool the system.
top
|