AUM System, Inc.

Home | Client Login | Employee Login | Sitemap | Downloads


Home > Products > Voice Print Technologies

Voice Print Technologies

Just as finger printing is an important biometric for authentication we can also use voice printing as a form of biometric for identification. Just as finger printing is a physiological biometric likewise voice printing uses unique information about one’s vocal tract and behavior of one’s speaking pattern for authentication. AUM has conducted extensive research in this area and has developed a speaker verification engine that has many uses in applications such as site access, forensics, smart cards, home incarceration, credit card authorization, secure phone access to banking, database services, and access to secure equipment.

Speaker recognition is divided into two areas: speaker identification and speaker verification.


Speaker Identification:

Speaker identification determines which registered speaker provides a given utterance from amongst a set of known speakers. For example, suppose we have a database of say 50 known speakers whose voice profiles have been registered then the goal is to select the closest speaker in the database to the one that is speaking and then depending on the set thresholds one speaker is selected as the matching one. Incase there is no close match then the voice identified is considered to be out of set. Most speaker identification systems are text independent, meaning a speaker can speak any word of his or her choice for identification purposes however the accuracy will decline with large sets of speakers and text independent speaker identification systems are optimal for smaller user sets, say 50 speakers.

In brief the speaker identification process is as follows:

  • Speech input
  • new voice print
  • extract features
  • compare against stored voice prints (1..N)
  • rank closest voice prints
  • select closest id if it meets the set threshold or select none in case of identification outside of set.


Speaker verification

In speaker verification we have a speaker who claims that he is Mr. X. The goal then is to compare claimant’s voice against Mr X’s preregistered voice profile. If the voice profiles are matching then it is said that the speaker was verified. In case there is a mismatch in the profiles then the claimant’s identity is said to be rejected. The criteria for acceptance and rejection of the claimed identity are a function of the features information that was extracted and was stored and features extracted from the new voiceprint. The result in speaker verification is a yes or no decision whereas in speaker identification the result is the identified speaker’s name. Unlike speaker identification systems speaker verification works best with text dependent text. For instance, a user might be required to speak his password to authenticate his claimed identity.

In brief the speaker verfication process is as follows::

  • Enter pin/userid
  • new voice print
  • extract features
  • compare against stored voice print of claimed speaker
  • Accept of Reject


Request Report on Speaker Verification


Contact Us Priority

AUM selected as a supplier by a Fortune 500 insurance company


About Us | Products | Collaborations | Services | Partners | Careers

Copyright ©1990-2010 AUM Systems, Inc. All rights reserved.

Home | Sitemap | Contact Us