| 
technologies
that can work alongside Prosody, a biometric SVI option is available.
SVI provides the most cost effective, convenient and secure way to access
personal information over the telephone. It enables applications to be
developed to provide access to sensitive data or enhances existing access
to secure areas.
The biometric technology captures the physical characteristics of the
human voice to verify individuals are ‘who they say they are’ using
just a single utterance over the telephone or a network connection.
Aculab SVI uses statistical information to confirm a caller’s
identity based on the characteristics of his or her unique voiceprint.
The availability of this host based SVI module brings yet more choice
to developers using Prosody media processing resources. Available under
a cost free licence, this confirms Aculab’s commitment to offer
unmatched value and features for speech processing.
The use of Aculab SVI can enhance a range of applications. For example,
it can provide secure access to a company’s Intranet/extranet
or corporate applications and reduce costs associated with resetting
PIN numbers.
Contact centre applications using SVI can reduce the workload of operators
and agents where previously verification of identity or password resets
required manual intervention. Agents can then concentrate on other revenue
generating activities.
SVI applications can be targeted at the telecoms sector for calling
card, cellular roaming, unified messaging and auto-attendant systems
to reduce fraud, protect personal information and gain competitive advantage.
Within the travel industry, SVI systems can be provided for frequent-customer
service interactions, promoting convenience and personalisation.
Simultaneous speech recognition and speaker verification allows applications
to verify the caller based on their voice and on a piece of personal
knowledge, for an even more secure verification. SVI based applications
can automatically detect callers whose voiceprints are difficult to
correctly authenticate and request additional information from them.
SVI is delivered as part of Aculab’s ASR and this allows SVI and
speech recognition to run simultaneously. Integration with the same server
software also minimises latency and reduces call duration. It has been
integrated seamlessly into the existing generic API.
With Prosody S or high density Prosody cards providing multiple channel
feeds, verification channel capacity is only limited by the processing
power of the ASR/SVI server. Using additional servers will expand the
channel capacity. Flexibility is built into the architecture, allowing
single or multiple distributed servers to fully exploit the capabilities
of the Prosody media processing resources.
Used in isolation, Aculab SVI is language independent. If used in conjunction
with ASR, it will support British and American English, North American
Spanish, French, Italian, and German. Each language is available on
a per verification basis, although they may not be used simultaneously
for a single recognition.
|
|
Prosody options
Aculab SVI is used with Prosody cards in PCI or cPCI formats that
provide an audio feed via an E1 or T1 network line interface.
It can also be used with Prosody S as the SVI software is compatible
with any media processing firmware that supports 64kbits/s record.
This allows ASR and SVI to be run in parallel with other media
processing resources, such as playback, echo cancellation and
DTMF detection, providing a versatile mixture of speech technologies
on the same platform.
Benchmarking
The frequency with which target speakers are rejected and the frequency
with which impostors are accepted are factors used to determine
the performance of a verification system. The system performance
is improved by reducing both values. It is possible to set a
threshold where the percentage of false rejections and false
acceptances are of the same value; this is called the ‘equal
error rate’ (EER).
Aculab SVI has achieved a 4% EER. This was achieved using single
utterance testing. A lower EER may be achieved if the application
uses pass phrases.
In terms of channel count, using a 1.9GHz P4 CPU with 400MHz bus
and 1Gb of RAM, SVI can verify 120 concurrent callers.
|
-
Speaker enrolment and deletion enables a new speaker to be added
to the system by providing some training data and a unique user name.
Additional data, plus a PIN or password for added security and further
training data, can be requested.
-
Speaker verification allows a speaker to make an identity claim
and the system accepts or rejects that claim.
-
Speaker identification will return the details of the speaker that
best matches the incoming speech.
-
Speaker adaptation can be used every time a speaker passes a subsequent
verification test; their template can be updated using the latest
examples of their speech.
-
High verification accuracy is achieved using only a few seconds
of speech.
-
Continuous monitoring can be used to monitor a speaker's voice
throughout a call. For example, if run concurrently with ASR to
recognise a password and verify the speaker, this enables a more
secure service provision.
-
Scalable architecture is possible where a distributed server allows
the system to be easily expanded.
-
Aculab’s generic API ensures ease of integration and
faster time to market.
-
A cost free licence enables SVI software to be downloaded from
software downloads, when and where you need it, with no per channel
costs or recurring licence fees.
-
Grunt detection is used to identify the endpoints of the utterance
and minimise system loading. Parameters can be adjusted via the
API.
-
Echo cancellation can be performed simultaneously on the Prosody
media processing resource card(s) to prevent spoken passwords being
echoed and leading to a false rejection.
-
Leverage existing infrastructure with SVI using any ordinary
telephone without the need for additional equipment such as
scanners or cameras.
-
Limit agent interventions by eliminating the need for manual
password resets. The system frees valuable agent time and can
pay for itself in a matter of months.
-
Reduce fraud while providing an effective means for secure voice-enabled
commerce. Cellular access and credit card information are just
two areas that can be protected with voiceprints for improved
security and consumer convenience.
-
Caller convenience can be increased when used in conjunction
with ASR and other recognisers i.e., the caller can be recognised
and verified as an account number is spoken, reducing the requirement
to remember and enter a password.
-
Enhanced security is gained when combined with a password; an
SVI system adds another level of security, confirming the correct
individual said the correct password.
|
|
